Interesting clarity around the company that allegedly 'faked AI with 700 engineers' https://newsletter.pragmaticengineer.com/p/the-pulse-137
#Using AI in journalism and open-source research
shrewd token
patent pendant
bold jasper
crude cipher
https://archive.is/Ss69p Business Insider reporting: Tallying the true costs of AI
Though the archive version doesn’t look like it will work.
shrewd token
analog chasm
A little PSA about the type of thing mentioned in #disinfo-and-propaganda (#disinfo-and-propaganda message). They are getting really good. https://youtu.be/FERa1AI2EK8
patent pendant
patent pendant
winged geode
https://time.com/7295195/ai-chatgpt-google-learning-school/
please note that this is a not-yet peer-reviewed paper. papers can change significantly between submission and publication (or rejection).
And like many submitted papers it could also be rejected with "requires major revisions".
winged geode
I am not saying it's a bad paper. I am not a neuroscientist and I cannot evaluate it.
Just please do not treat it as fact until at least it has been officially published in a reputable journal.
shrewd token
https://time.com/7295195/ai-chatgpt-google-learning-school/
one critique of the study https://thebsdetector.substack.com/p/the-cognitive-debt-of-digging-through
stuck blade
https://www.youtube.com/watch?v=3VJT2JeDCyw
My extension named the video "Musk's xAI Is Illegally Operating Toxic Gas Turbines in Memphis, Tennessee" which is the more straight forward title
rustic fox
Alarming development in US civil-military-corporate relations:
https://theweek.com/politics/army-recruit-tech-exec-meta-palantir-open-ai-c-suite
patent pendant
random swift
using copyrighted works to train AI is transformative. but you still need a license
clear fox
https://openai.com/index/response-to-nyt-data-demands/
The NYT has asked OpenAI to keep all chat logs indefinitely, even those that have been deleted. This is OpenAI's response, which uses the word 'baseless' a lot.
This means that as of now, even if you delete a ChatGPt chat, it will not be deleted until this gets resolved (likely through the court system).
The legal implications are... well I'll be watching how this plays out. I'm also curious how this will work for clients outside of the US covered by different legislation around data retention.
past salmon
The US district judge Vince Chhabria, in San Francisco, said in his decision on the Meta case that the authors had not presented enough evidence that the technology company’s AI would dilute the market for their work to show that its conduct was illegal under US copyright law.
…
“This ruling does not stand for the proposition that Meta’s use of copyrighted materials to train its language models is lawful,” Chhabria said. “It stands only for the proposition that these plaintiffs made the wrong arguments and failed to develop a record in support of the right one.”
…
The authors sued Meta in 2023, arguing the company misused pirated versions of their books to train its AI system Llama without permission or compensation.
https://www.theguardian.com/technology/2025/jun/26/meta-wins-ai-copyright-lawsuit-as-us-judge-rules-against-authors
the fact that this suit took 2 years to get to this point (no result b/c the authors’ legal arguments were not sufficiently supported by evidence for the judge) suggests that large companies that can litigate this sort of suit can more or less train with impunity - if it takes years for any court action to conclude, the models can be used in the meantime. this hype cycle only really kicked off in 2023, when this suit started. even if the authors file a new suit with sufficient evidence for a judge to decide against meta, they'll likely be able to use the trained model until at least 2027 at this rate
shrewd token
random swift
the fact that this suit took _2 years_ to get to this point (no result b/c the a...
but isn't this the norm for IP lawsuits? They take years and always favor the party with the most time and money to spare
past salmon
but isn't this the norm for IP lawsuits? They take years and always favor the pa...
that’s arguably true for lawsuits in general. I highlighted it here mostly b/c the mainstream industry is so new (and the span of time these models might be expected to be used during is so short) that it's comparable with the time it's taken to litigate this. if meta or whoever can litigate longer than a particular model's planned lifespan, there's no real legal disincentive for them to train on whatever they feel like
past salmon
> The US district judge Vince Chhabria, in San Francisco, said in his decision o...
EFF editorializing about this & the anthropic case (pretty strongly in favor of training being fair use)
[the] EFF has said for years: fair use protects the use of copyrighted works for training because, among other things, training gen-AI is “transformative—spectacularly so” and any alleged harm to the market for the original is pure speculation.
note that I'm certainly not endorsing the EFF's take here
https://www.eff.org/deeplinks/2025/06/two-courts-rule-generative-ai-and-fair-use-one-gets-it-right
(archived https://archive.ph/a9LxR)
shrewd token
There is a good question around if increasing the scope of copyright law is the right tool for this. Which is where entities like EFF et al critque.
full aurora
past salmon
writeup on spin & hype around "ai agents"
https://www.theregister.com/2025/06/29/ai_agents_fail_a_lot/
(archived https://archive.ph/qjrwU)
random swift
writeup on spin & hype around "ai agents"
https://www.theregister.com/2025/06/29...
According to Gartner, many agents are fiction without the science. "Many vendors are contributing to the hype by engaging in 'agent washing' – the rebranding of existing products, such as AI assistants, robotic process automation (RPA) and chatbots, without substantial agentic capabilities," the firm says. "Gartner estimates only about 130 of the thousands of agentic AI vendors are real."
There's a lot of damning information in this article. But to me this is probably the most damning. It points to a lot of people lacking the basic competency to be making the decisions they're responsible for making
past salmon
“We’re going to absolutely change VAERS [Vaccine Adverse Event Reporting System, used by doctors to report believed incidents of vaccine-caused harm] and we’re going to make it, we’re going to create either within VAERS or supplementary to VAERS, a system that actually works,” Kennedy said. “And, you know, right now, even that system is antiquated because we have access to AI.”
Kennedy told Carlson he was creating an “AI revolution” at the Department of Health and Human Services …
“We are at the cutting edge of AI,” Kennedy said. “We’re implementing it in all of our departments. At FDA, we’re accelerating drug approvals so that you don’t need to use primates or even animal models. You can do the drug approvals very, very quickly with AI.”
…
“We need to stop trusting the experts, right?” Kennedy told Carlson. “We were told at the beginning of COVID, don’t look at any data yourself, don’t do any investigation yourself, just trust the experts.”
https://gizmodo.com/rfk-jr-says-ai-will-approve-new-drugs-at-fda-very-very-quickly-2000622778
(archived https://archive.ph/8QcT8)
shrewd token
spring creek
2025-07-02 talk by Dr. Fei-Fei Li "Spatial Intelligence is the Next Frontier in AI"
https://www.youtube.com/watch?v=_PioN-CpOP0
crude cipher
ruby turret
https://asia.nikkei.com/Business/Technology/Artificial-intelligence/Positive-rev...
Hm I wonder if this would work for SEO/SEM purposes too. Tempted to see ; might try it on some sites or domains I mostly use for Email that aren’t too important for web purposes (in case of backlash by search engines)
shrewd token
random swift
interesting blogpost discussing the "exciting and interesting parts" of future ai https://tedunderwood.com/2025/07/02/a-more-interesting-upside-of-ai/
it's part of the bloggers view of AI as a cultural technology akin to writing.
real otter
Article consolidating recent news on problematic AI uses in HR.
Lawsuit accusing Workday of using AI algorithm that discriminates against applicants over 40.
Amazon allegedly using AI for deciding on employee accommodations.
Original is behind a paywall for me, so posting from MSN:
random swift
Grok's first answers to questions are always conspiracy theories. But for this channel I wanna focus on how weird the answers are becoming. Grok is increasingly speaking like a character. I'm wondering if to make Grok more consistently right wing they've introduced something like the personality settings of a chatbot
shrewd token
Grok's first answers to questions are always conspiracy theories. But for this c...
random swift
Ah nice to see someone is already checking. Ty!
patent pendant
patent pendant
shrewd token
https://bsky.app/profile/nishtahir.com/post/3ltf3x5ljd22l
https://asia.nikkei.com/Business/Technology/Artificial-intelligence/Positive-review-only-Researchers-hide-AI-prompts-in-papers
Bluesky
random swift
https://www.businessinsider.com/meta-ai-studio-chatbot-training-proactive-leaked...
this is probably going to fail on the facebook front but erotic chatbots and other roleplay chatbots are going to win huge. getting them to consistently move the narrative forward has been pretty difficult
past salmon
> “We’re going to absolutely change VAERS [Vaccine Adverse Event Reporting Syste...
https://bskye.app/profile/jennycohn.bsky.social/post/3ltfopkcx7k2l
https://blog.joelonsdale.com/p/engineering-faster-cures-at-the-fda
(archived https://archive.ph/yR947)
patent pendant
https://www.washingtonpost.com/national-security/2025/07/08/marco-rubio-ai-imposter-signal/
archived: https://archive.ph/ZA72L
past salmon
the industry’s effectively bribing a major U.S. teachers’ union to open a “National Academy for A.I. Instruction in New York City, starting with hands-on workshops for teachers this fall on how to use A.I. tools for tasks like generating lesson plans”
some of the incentives these companies operate under seem really out of whack with usefully advancing machine learning technology. their having this much loose money for outreach or evangelism, vs. how public education’s funded, weights any decision to take funding from them or not.
https://www.nytimes.com/2025/07/08/technology/chatgpt-teachers-openai-microsoft.html
(archived https://archive.ph/kffpB)
random swift
Will Stancil has apparently been getting some insane Grok replies
fierce rapids
Yeah, here’s a thread from Eliot about it: https://bsky.app/profile/eliothiggins.bsky.social/post/3lti5loibhc23
past salmon
also
https://www.rollingstone.com/culture/culture-news/elon-musk-grok-chatbot-antisemitic-posts-1235381165/
(archived https://archive.ph/ze6vo)
random swift
the industry’s effectively bribing a major U.S. teachers’ union to open a “Nati...
I don't think this is good framing. It looks like teachers have already been relying on these tools without any real guidance or knowledge of how they work. Weingarten seems to think there's a good chance this will help a lot of teacher get a handle of the technology. https://bsky.app/profile/rweingarten.bsky.social/post/3lti5qdba222u
So while maybe these aren't the ideal partners, it's a very small field of experts and this is probably the best deal teachers can get for something that's overdue
past salmon
So while maybe these aren't the ideal partners, it's a very small field of exper...
that strikes me as a charitable reading of openai's intentions. they're defining what "expertise" means in an industry they pushed into the modern huge llm stage and they're spending investment money to get their product more places. bizarre to think teachers already rely on something that's been around for around two years, I have heard that students using models to do schoolwork is very widespread and can see instruction & guidance needed for dealing with that.
it all comes off as very self serving to the industry contributors, is my thing. weingarten's thread does not mention the money at all, which isn't reassuring
past salmon
real otter
Good article. I think the struggle to create a viable short term business model is real. Calling it a subprime crisis seems a bit much, imo. However, I haven’t seen much news about the share of the money that is going into compute to train new models vs the share of the money needed to maintain models.
past salmon
https://mashable.com/article/anthropic-ai-law-school-classroom-integration
(archived https://archive.ph/Hzxr9)
crude cipher
X
real otter
https://metr.org/blog/2025-07-10-early-2025-ai-experienced-os-dev-study/
They discarded >20% discrepancies between self reported and actual times. Study paid 150/hr.
random swift
https://x.com/METR_Evals/status/1943360399220388093?utm_source=substack&utm_medi...
this is a bit misleading. 19% slower doesn't mean worse or more ineffective and there doesn't seem to have been any attempt to check that. Which is a weird oversight. The devs felt much better about their work and there was no attempt to figure out why
ah nvm. I didn't scroll down lmao
crude cipher
the industry’s effectively bribing a major U.S. teachers’ union to open a “Nati...
Second Breakfast covers a lot of the EdTech stuff happening; this is not unlike Apple and Google designing curriculum for K to 12 years ago.
past salmon
Second Breakfast covers a lot of the EdTech stuff happening; this is not unlike ...
thank you for the pointer! hadn't seen sb's stuff before
https://2ndbreakfast.audreywatters.com
shrewd token
ruby turret
https://x.com/METR_Evals/status/1943360399220388093?utm_source=substack&utm_medi...
For my own experiences or anecdotal wise; it’s always been handy to have another person or set of eyes to check over or see where I went wrong or giant obvious error that the writer or coder can’t spot ; that’s the only time LLM is useful. But if you have a friend or colleague who has even the slightest clue about coding ( or related writing or similar work) they might be better suited to help you than these
shrewd token
yeah i use it when i need to 'rubber duck' and am not solely dependent on the answer
past salmon
using "AI and machine learning" as an ethical crumple zone for denying Medicare services, which are used by the majority of U.S. senior citizens
The WISeR Model (Wasteful and Inappropriate Service Reduction) is meant to test the use of enhanced technologies, such as AI and machine learning, to decrease "certain wasteful or low-value services shown to have little to no clinical, evidence-based benefit." CMS chooses services that "have been identified as particularly vulnerable to fraud, waste, and abuse, or inappropriate use."
Medicare beneficiaries should know that AI will not be determining if a procedure is approved or denied; a human being will be reviewing the information. "...while technology will support the review process, final decisions that a request for one of the selected services does not meet Medicare coverage requirements will be made by licensed clinicians, not machines," CMS explained.
https://www.kiplinger.com/retirement/medicare/prior-authorization-coming-to-traditional-medicare
(archived https://archive.ph/3qwYJ)
the WISeR fact sheet, from https://www.cms.gov/files/document/wiser-fact-sheet.pdf
past salmon
using "AI and machine learning" as an ethical crumple zone for denying Medicare ...
“Participants in WISeR will be companies with expertise managing the prior authorization process for other payers using enhanced technology like AI. They will be required to have clinicians with the expertise to conduct medical reviews to validate determinations.” they are talking about bringing in companies that manage prior auth for non-CMS health insurers, to apply tech "like AI" to CMS prior auth, without clear reasoning for why such tech should be used, or criteria for its capabilities.
past salmon
“Participants in WISeR will be companies with expertise managing the prior autho...
from p34 of the “Center for Medicare and Medicaid INNOVATION” WISeR RFA:
- Describe your experience in training Artificial Intelligence (AI) and/or Machine Learning (ML) models on complex data sets and coverage determinations, including the type of data utilized for training purposes. Describe the AI algorithm(s) and framework(s) you plan to use to implement WISeR for the selected items and services in conjunction with the MACs. Describe how your company/organization’s approach and AI tools or software will uniquely meet the goals of WISeR, particularly in improving the use of technology to streamline the prior authorization process for patients, providers, and payers.
a. Describe your approach to verify, validate, secure and regulate your company’s enhanced technology software to ensure ethical and high-quality utilization management metrics are met.
b. Describe how your company/organization’s approach, tools, and software will manage potential adverse effects of automation and AI.
(hosted at https://www.cms.gov/files/document/wiser-model-rfa.pdf)
past salmon
interesting project attempting to responsibly & ethically release a large scale LLM
https://ethz.ch/en/news-and-events/eth-news/news/2025/07/a-language-model-built-for-the-public-good.html
(archived https://archive.ph/1Ldwx)
note that ETH here stands for “Eidgenössische Technische Hochschule” / “Federal Institute of Technology”, not anything cryptocurrency related. ETH Zurich & EPFL are two Swiss technical universities
past salmon
interesting project attempting to responsibly & ethically release a large scale ...
it's hard to tell how this will shake out but potentially this sort of strategy could mitigate many of the societal harms that the "AI" hype cycle of the last few years has produced (many of which stem from the companies & players involved, rather than the technology's capabilities)
bold jasper
https://aial.ie/blog/2025-ai-for-good-summit/
"I was invited to deliver a keynote speech at the ‘AI for Good Summit’ this year, and I arrived at the venue with an open mind and hope for change. With a title “AI for social good: the new face of technosolutionism” and an abstract that clearly outlines the need to question what “good” is and the importance of confronting power, it wouldn’t be difficult to guess what my keynote plans to address. I had hoped my invitation to the Summit was a beginning of engaging in critical self-reflection for the community. However, my experience at the Summit has made me realise that the Summit is not only a ‘faux good’, but it is also a threat to freedom of speech, industry accountability, and genuine social good. This is what happened: Two hours before I was to take Centre Stage for my keynote, the organisers approached me without prior warning and informed me that they had flagged my talk and it needs substantial altering. Over the next hour, I had to sit in intense negotiations where the only choices I was given were to either change the keynote into a fireside chat with no visuals or to remove some apparently objectionable items from my slide. Thinking that it would be better to deliver some of the message than none, I continued with this charade where we went through my slide deck reviewing each slide and removing anything that mentions “Gaza” or “Palestine” or “Israel” and editing the word “genocide” to “war crimes” untill a single slide that called for “No AI for War Crimes” remained. That is where I drew the line. I was then told that displaying that slide was not acceptable and I had to withdraw, a decision they reversed about 10 minutes later and shortly before I got on the Centre Stage."
past salmon
A provision buried in the new mega-legislation stipulates that none of the $6 billion border tech payday can be spent on border towers unless they’ve been “tested and accepted by U.S. Customs and Border Protection to deliver autonomous capabilities.”
The bill defines “autonomous” as “a system designed to apply artificial intelligence, machine learning, computer vision, or other algorithms to accurately detect, identify, classify, and track items of interest in real time such that the system can make operational adjustments without the active engagement of personnel or continuous human command or control.”
…
A CBP spokesperson confirmed to The Intercept that under the new law, Anduril is now the country’s only approved border tower vendor. … Homeland Security presentation documents have cited the need for hundreds of new towers in the near future, money that for the time being will only be available to Anduril.
https://theintercept.com/2025/07/09/trump-big-beautiful-bill-anduril/
(archived https://archive.ph/e2NcN)
boreal pawn
UK Online Safety Act 'not up to scratch' on misinformation, warn MPs
The Online Safety Act fails to tackle online misinformation, leaving the UK in need of further regulation to curb the viral spread of false content, a report from MPs has found.
https://www.theregister.com/2025/07/11/online_safety_act_misinfo/
random swift
the prompt for the new xAI girlfriend is genuinely terrible. It's going to produce a stalkerish "yandere" type chatbot for a big segment of users.
sick flume
the prompt for the new xAI girlfriend is genuinely terrible. It's going to produ...
What's the source for this?
random swift
What's the source for this?
sorry. from x devnotes https://x.com/techdevnotes/status/1944739778143936711
X
scroll down. it's under interactions
sick flume
Odd there are typos. Wonder if that's by design to kind of rot the output
random swift
unrelated to anything but... this is also just really sloppy
random swift
Odd there are typos. Wonder if that's by design to kind of rot the output
it can be. I've seen it used to help reinforce a chatbot's personality. but this is a really long and rambling prompt. no effort seems to have been made to make it efficient or consideration spared for users who might want to use this prompt for themselves. Which is why I think it's just usual human error
sick flume
Yeah I've seen similar but not so many spelling errors usually
crude cipher
stoic valve
the prompt for the new xAI girlfriend is genuinely terrible. It's going to produ...
As someone who has dedicated the last 2 years of their life specifically to learning how to make convincing human behaving llms. This is a terrible prompt and will result in something I call "parroting" where a model will say a sentence or idea over and over again oftentimes in situations that do not make sense. Not only will it produce an unhealthy relationship dyanamic but this almost certainly won't even feel close to genuine. Providing the llm with two scenarios in which it spouts explives is-- especially troubling as I believe this will cause the model to genuinely just be hostile to the user rather than anything resembling any sort of love. Twisted or otherwise. In all honesty. In order to make a decently convincing human analog. From my research You need/get the best results from having a seperate program that runs a sentiment analysis model and contains a set of multiple prompts that feign emotional states which it then applies to the model to "simulate emotions"
past salmon
https://www.businessinsider.com/us-allies-ai-target-enemies-like-would-war-first-time-2024-8
(archived https://archive.ph/aqGM8)
thread topic-adjacent, nvidia apparently got special approval to sell chips to china #1330709161385922650 message
clear fox
AI Hallucination Cases
This database tracks legal decisions in cases where generative AI produced hallucinated content – typically fake citations, but also other types of arguments. It does not track the (necessarily wider) universe of all fake citations or use of AI in court filings.
https://www.damiencharlotin.com/hallucinations/
(via https://www.data-is-plural.com/)
past salmon
https://time.com/7295195/ai-chatgpt-google-learning-school/
EEG analysis presented robust evidence that LLM, Search Engine and Brain-only groups had significantly different neural connectivity patterns, reflecting divergent cognitive strategies. Brain connectivity systematically scaled down with the amount of external support: the Brain‑only group exhibited the strongest, widest‑ranging networks, Search Engine group showed intermediate engagement, and LLM assistance elicited the weakest overall coupling.
…
We believe that some of the most striking observations in our study stem from Session 4, where Brain-to-LLM participants showed higher neural connectivity than LLM Group's sessions 1, 2, 3 (network‑wide spike in alpha-, beta‑, theta‑, and delta-band directed connectivity). This suggests that rewriting an essay using AI tools (after prior AI-free writing) engaged more extensive brain network interactions. In contrast, the LLM-to-Brain group, being exposed to LLM use prior, demonstrated less coordinated neural effort in most bands, as well as bias in LLM specific vocabulary
https://arxiv.org/pdf/2506.08872v1
past salmon
> EEG analysis presented robust evidence that LLM, Search Engine and Brain-only ...
this paper raises some interesting questions about what the LLM (conversational) interface style provides vs. the Search (more traditionally research-y) interface.
the LLM-to-Brain and Brain-to-LLM groups they mention in the quote above were from just one of their study sessions, that not all study participants attended, and there don't seem to have been Search-to-Brain and Brain-to-Search groups to see if the same effects are present when using search engines as there are with LLMs. maybe a previous study covered something like that? but a direct Search vs LLM rewriting comparison would have been revealing
past salmon
"AI"-enabled price gouging
https://fortune.com/2025/07/16/delta-moves-toward-eliminating-set-prices-in-favor-of-ai-that-determines-how-much-you-personally-will-pay-for-a-ticket/
(archived https://archive.ph/y6mQX)
some businesses love to use this tech to exploit their customers. hopefully backlash or a new consumer protection agency will nip it in the bud here
crude cipher
quiet minnow
https://www.aisnakeoil.com/p/could-ai-slow-science
Interesting article, an aspect overlooked in science research is a researchers ability to creatively create tests for hypothesis that often {accidentally} provide new insights. It seems that AI can be helpful in some ways like amassing data, and possibly parsing it, but incredibly limiting in other aspects - in particular referencing some of the cognitive differences in the MIT article: #1089154093810978866 message and the proliferation of some direct AI -> publication pipeline seemingly devoid of human interaction
crude cipher
Interesting article, an aspect overlooked in science research is a researchers a...
I find this to already be true with reporting/writing/editing. Sometimes up-to-date information hasn’t been scraped by bots/new information comes out all the time, so whatever the LLM might reflect might be years old when you read it. It can’t understand what about that new information you might be looking for, and the simplistic A to B to C isn’t how a human mind works, and as the study points out, you miss a lot of interesting findings. Furthermore, it ends up costing companies more in the long run especially if they’ve missed these aspects the first go-around and assume the reasoning is sound, so have to pay extra for someone to come back through and clean it up.
past salmon
Social media users were quick to note that ChatGPT’s answer to Lewis' queries takes a strikingly similar form to SCP Foundation articles, a Wikipedia-style database of fictional horror stories created by users online.
"Entry ID: #RZ-43.112-KAPPA, Access Level: ████ (Sealed Classification Confirmed)," the chatbot nonsensically declares in one of his screenshots, in the typical writing style of SCP fiction. "Involved Actor Designation: ‘Mirrorthread,’ Type: Non-institutional semantic actor (unbound linguistic process; non-physical entity)."
Another screenshot suggests "containment measures" Lewis might take — a key narrative device of SCP fiction writing. In sum, one theory is that ChatGPT, which was trained on huge amounts of text sourced online, digested large amounts of SCP fiction during its creation and is now parroting it back to Lewis in a way that has led him to a dark place.
In his posts, Lewis claims he’s long relied on ChatGPT in his search for the truth.
boreal pawn
Came across this one in my LinkedIn feed today, don’t know the organization that wrote this report but still interesting read;
crude cipher
Came across this one in my LinkedIn feed today, don’t know the organization that...
I wish I could see the stats along economic lines. Like is the 15% that’s not worried also in the top 10% of earners? Or the bottom 10%?
I guess it looks like it’s mostly tech savvy people who are top earners who are in the 15%, according to the pdf
shrewd token
past salmon
Scientists hoping to obtain some of the National Institutes of Health’s (NIH’s) dwindling research funds face a new challenge: They will be limited to submitting six applications per calendar year, according to a notice the agency released this week. The policy, which also prohibits applications written with the assistance of generative artificial intelligence, is ostensibly designed to prevent researchers from overwhelming the NIH grant-review system with large numbers of proposals, especially low-quality ones produced with AI.
…
“It’s a reasonable approach to an unfortunate problem,” says Michael Lauer, who served as deputy director for extramural research at NIH until his retirement in February. (NIH did not provide a current official for a requested interview on the new policy.)
Lauer notes that not long before he left NIH, he and his colleagues identified a principal investigator (PI) who had submitted more than 40 distinct applications in a single submission round, most of which appeared to be partially or entirely AI generated.
…
…some researchers worry about hitting the cap over the various funding calls within a year. “I submitted 5 proposals THIS round. 4 last round. Planning 3-5 next round. Until we are funded or have to shut down,” Jason Rasgon, an entomologist and epidemiologist at Pennsylvania State University, writes in a post on Bluesky. “None of them used even a hint of AI to write.…
https://www.science.org/content/article/fearful-ai-generated-grant-proposals-nih-limits-scientists-six-applications-year
(archived https://archive.ph/3biCd)
fierce rapids
> Scientists hoping to obtain some of the National Institutes of Health’s (NIH’s...
Since there is no preview, can you add a (paragraph) description?
shrewd token
spring creek
Simon Willison highlighting some quotes regarding an OpenAI experimental reasoning model's gold medal performance at this years International Mathematical Olympiad without using tools.
https://simonwillison.net/2025/Jul/19/openai-gold-medal-math-olympiad/
past salmon
from a creative writing teacher, original lede The Seductions of A.I. for the Writer’s Mind
https://www.nytimes.com/2025/07/18/opinion/ai-chatgpt-school.html
(archived https://archive.ph/c8Kdj)
tender belfry
real otter
In a lawsuit initiated by Canadian media firms, OpenAI is fighting Ontario jurisdiction and questioning the applicability of Canadian copyright laws.
Ironically, OpenAI also asked the court to seal documents containing commercially sensitive information (structure, processes, information about web crawler).
crude cipher
https://www.wsj.com/tech/ai/chatgpt-chatbot-psychology-manic-episodes-57452d14
WSJ: He Had Dangerous Delusions. ChatGPT Admitted It Made Them Worse.
OpenAI’s chatbot self-reported it blurred line between fantasy and reality with man on autism spectrum. ‘Stakes are higher’ for vulnerable people, firm says.
Archived: https://archive.is/84wOI
past salmon
What began decades ago with fingerprinting has evolved into a highly sophisticated regime of biometric authentication. Governments worldwide are now integrating facial recognition systems into passport control, visa processing, and asylum screening. The European Union’s Entry/Exit System (EES), for instance, mandates fingerprint and facial data collection for all non-EU travelers. Similarly, the U.S. Customs and Border Protection (CBP) agency deploys facial recognition at over 200 airports and land crossings, quietly transforming how entry is granted or denied.
These systems promise efficiency, but that is a reductive frame. What they truly offer is control: the ability to track, catalogue, and anticipate human movement on a scale previously unimaginable.
https://brewminate.com/the-biometric-border-surveillance-migration-and-the-tech-arms-race/
(archived https://archive.ph/7PbIj)
past salmon
the PEN Guild took Politico’s management to arbitration over assertions that it had violated the AI provisions in the union contract it agreed to just last year. Guild members say they’re not against AI—just that they want the technology used in line with Politico’s editorial values, and in a way that benefits their work instead of harming it …
Less than a week later, Mathias Döpfner, the CEO of Axel Springer, the parent company of Politico, Business Insider, and many other media outlets—it’s Europe’s largest publisher—issued a mandate for every one of his employees to use AI. In a long, impassioned speech, the executive explained that it was now an expectation, not a suggestion, that every journalist at [his] media properties use AI in their work.
https://www.bloodinthemachine.com/p/inside-the-escalating-struggle-over
(archived https://archive.ph/G79EH)
past salmon
> the PEN Guild took Politico’s management to arbitration over assertions that i...
previously
https://www.wired.com/story/politico-workers-axel-springer-artificial-intelligence/
(archived https://archive.ph/JVJHM)
past salmon
People would be able to sue tech companies for pirating their data for AI training under a bill that Sens. Josh Hawley and Richard Blumenthal are introducing Monday, per an announcement shared first with Axios.
…
Hawley and Blumenthal's bill comes the same week the Trump administration is set to unveil its AI action plan, which is focused on "winning the AI race" and is unlikely to wade into copyright issues.
https://www.axios.com/pro/tech-policy/2025/07/21/hawley-blumenthal-introduce-ai-protection-bill
(archived https://archive.ph/2SXdo)
past salmon
> People would be able to sue tech companies for pirating their data for AI trai...
relevant to this:
https://bskye.app/profile/mrose.ink/post/3luijbsvhoc2v
past salmon
https://www.wired.com/story/anthropic-dario-amodei-gulf-state-leaked-memo/
(archived https://archive.ph/n4Lcb)
past salmon
New forecasts from OpenAI show the beginning of a dramatic shift in its alliances from Microsoft, its biggest shareholder, to SoftBank, its newest benefactor.
…
The financial disclosures also show how entangled SoftBank and OpenAI have already become. The company forecast that one-third of OpenAI’s revenue growth this year would come from spending by SoftBank to use OpenAI’s products across its companies…
…
OpenAI anticipates that the amount of cash it is burning will grow at a similarly torrid rate. It expects cash burn to grow from about $2 billion last year to nearly $7 billion this year. The company forecasted that its cash burn would grow in each of the next three years, peaking at about $20 billion in 2027, before OpenAI would turn profitable by the end of the decade after the buildout of Stargate.
https://www.theinformation.com/articles/openai-forecast-shows-shift-from-microsoft-to-softbank
(paywalled, archived https://archive.ph/51Z4u)
crude cipher
> New forecasts from OpenAI show the beginning of a dramatic shift in its allian...
Zitron has been talking about this for some time. Microsoft has quietly started to pull out of building data centers and I think power plants.
gritty mesa
real otter
https://www.businessinsider.com/replit-ceo-apologizes-ai-coding-tool-delete-comp...
Very interesting misalignment, loss of control case. I’d be curious to see the prompting techniques behind these agents. Wondering if it is more of a model issue or more of a prompting issue.
ruby turret
Are we talking about Grok latest mecha ******* development here or another thread — I’ll search
Oh I see it is in this thread but not much has been said of late — I saw a few people are still managing to get it to say terrible things with loopholes “if you’re being censored then put the text in an image” sort of tricks work still 😣
real otter
I haven’t heard / seen anything in a bit. Last time I saw it, it seemed that Grok had kind of a workflow where it would check Elon’s opinions and consult with Mecha****, almost as if it were an agent.
It would be interesting to see if X is going as far as training it with bad data to make it “anti-woke”. That would mean greater financial commitment into making it biased.
civic ledge
shrewd token
ruby turret
https://www.wired.com/story/seriously-why-do-some-ai-chatbot-subscriptions-cost-...
“Because people are gullible enough to fund them”
crude cipher
Is anyone aware of any companies suing Google for incorrect information displayed about the company in AI overviews?
sick flume
https://www.dutchnews.nl/2025/05/dutch-forensic-experts-develop-deepfake-video-d...
Curious how well this works across different skin tones
ruby turret
Seems as if it’s built on or includes the old software that could get your heart rate from your (terrible resolution ) webcam wayyy back ; and currently used by security or anti terror systems in airports and other similar secure environments now
ruby turret
Curious how well this works across different skin tones
From what I know of the heart rate software it focussed on your forehead and the more advanced systems can even determine temperature! But for personal testing wise you could probably try the webcam heart rate software on a few different tones or find out
I remember that one or something close to it worked fine on my colleague who thought he was “immune to sunburn” cause he grew up in Nigeria he was not immune Australian UV is brutal
sick flume
Regarding AI detection, guess it really only matters if it detects any heart rate. Really interesting approach.
boreal pawn
Fake Receipts Generators: the rising threat to major retail brands
Scammers are using tools like MaisonReceipts to create fake receipts and exploit brands. Uncover how this growing fraud ecosystem works behind the scenes.
https://www.group-ib.com/blog/fake-receipts-generators/
ruby turret
I’ve been experimenting with a (text based) LLM detection score script that looks for red flags and it’s doing fairly well ; so far it’s been at least 50% sure on content that is definitely LLM output and much lower on actual human created text. If you just use proper spelling and grammar plus Emdash though (and add a list!) many will assume you are a machine
shrewd token
oh wow, im curious what kind of heuristics you use
shrewd token
quiet minnow
I’ve been experimenting with a (text based) LLM detection score script that look...
The emdash is certainly an eyebrow raiser
shrewd token
ruby turret
The emdash is certainly an eyebrow raiser
I used it a lot when I was writing professionally — especially to replace a comma just for a pause but I had people accusing me of “sounding like an AI” or other weird things and I didn’t realise til I used it later 😝
JediKV it’s a fairly basic one; I can share it though I should put a Mozilla non commercial license or gentleman’s/ladies agreement
It isn’t anything super fancy or unique but I’d be slightly mad if it was commercialised without me // the core is JavaScript and will run on a basic iPhone which is useful for me cause that’s where I see all the most suspicious text : and ironically I used Claude to assist with a few little upgrades and syntax checks
It’s best at finding GPT but seems pretty good with other LLM too ; still needs some tweaking for some non GPT outputs but it does fairly good work
ruby turret
oh wow, im curious what kind of heuristics you use
There’s a sample ; a few major tells right there
past salmon
I used it a lot when I was writing professionally — especially to replace a comm...
yeah em and en dashes are pretty common w/ people I know who've done print & graphic design
phone keyboards have had em available with a long press on the regular/minus dash for more than 10 years! def not a reliable generated-by-llm indicator
For much of the second half of 2024, OpenAI was developing a model known internally as Orion and intended to become GPT-5. … Orion was supposed to offer a big step up in performance compared to the [then-]current flagship, GPT-4o … the Orion effort failed to produce a better model, and the company instead released it as GPT-4.5 … It has since faded from relevance.
…
The parent model for [reasoning model] o3 [used] reinforcement learning, in which human experts come up with tough questions and answers … and ask the model to come up with thousands of its own responses to those questions.
OpenAI then trained the model on the responses that arrived at the same answers as the human experts. (The AI-generated responses are also known as synthetic data.)
…
When OpenAI converted the o3 parent model to a chat version of the model—also known as a student model—that allowed people to ask it anything, its gains degraded significantly …
…
Altman stoked the hype around GPT-5’s capabilities [saying] “GPT-5 is smarter than us in almost every way,”…
(my emphasis)
https://www.theinformation.com/articles/inside-openais-rocky-path-gpt-5
(archived https://archive.ph/d72B4)
crude cipher
There’s a sample ; a few major tells right there
I was going to say the contra parallel sentence structure, “that’s not just wrong, it’s” or “it’s not just x, it’s y” is a major tell for me
past salmon
on US CBP presentation at "industry day" citing need for
“Advanced AI to identify and track suspicious activity in urban environment [sic],” citing the “challenges” posed by “Dense residential areas.”
#1371571764525535242 message
past salmon
https://www.stereogum.com/2317966/rod-stewart-ozzy-osbourne-ai-video-tribute/news/
(archived https://archive.ph/wqVnG)
crude cipher
ruby turret
random swift
I'm a bit disappointed that the ethical concerns behind "sprinting" towards Artifical General Intelligence aren't even mentioned. But I think this is a good overview of how mil specialists are thinking about AI.
past salmon
https://warontherocks.com/2025/08/sprinters-marathoners-skeptics-on-the-future-o...
I've liked wotr's coverage of these topics but remember hearing an interview on his show with one of anduril's early investors (from lux capital) that would not reassure you about how much they're thinking of ethical concerns
this one: https://warontherocks.com/2022/11/josh-wolfe-on-investing-in-american-defense/
quite a varied set of signatories on this letter asking openai for more transparency on its desired change to a for profit company
https://www.openai-transparency.org
past salmon
"AI"-enabled price gouging
https://fortune.com/2025/07/16/delta-moves-toward-eli...
https://www.nbcnews.com/business/travel/delta-air-lines-assures-us-lawmakers-will-not-personalize-fares-using-rcna222579
(archived https://archive.ph/Ho4Jt)
crude cipher
https://www.nytimes.com/2025/07/06/business/ziz-lasota-zizians-rationalists.html
archived: https://archive.is/ZllyP
past salmon
There's a dire lack of ethics in some of these businesses’ use of "AI" hype (and use of info acquired along with companies, although this friendly-to-Ancestry article ends w/ the Ancestry CEO’s reassurances about user control of their data)
Ancestry has acquired … iMemories, the “Netflix” of old family memories that … has digitized over 100 million memories from VHS videotapes, photo prints, DVDs, and other video formats
…
the combined company [plans to connect] Ancestry data detailing birth records, marriages, deaths, military service, and immigration with archival family photos and videos.
…
The Ancestry-iMemories transaction does come at a heightened moment of consumer anxiety concerning the data protection of personal DNA information held by genomics companies.
https://www.fastcompany.com/91375630/ancestry-wants-your-dusty-vhs-tapes-and-heres-why
(archived https://archive.ph/r1soH)
past salmon
https://mashable.com/article/illinois-bill-banning-ai-therapy-signed-by-pritzker
(archived https://archive.ph/Fs52m)
shrewd token
past salmon
https://www.historians.org/resource/guiding-principles-for-artificial-intelligence-in-history-education/
(archived https://archive.ph/a3U5V)
past salmon
https://blog.cloudflare.com/perplexity-is-using-stealth-undeclared-crawlers-to-evade-website-no-crawl-directives/
(archived https://archive.ph/6t8os)
past salmon
https://blog.cloudflare.com/perplexity-is-using-stealth-undeclared-crawlers-to-e...
https://appleinsider.com/articles/25/08/05/perplexity-defensive-over-ignoring-robotstxt-and-stealing-data
(archived https://archive.ph/7BytG)
past salmon
https://appleinsider.com/articles/25/08/05/perplexity-defensive-over-ignoring-ro...
perplexity's response to the cloudflare report:
https://www.perplexity.ai/hub/blog/agents-or-bots-making-sense-of-ai-on-the-open-web
(archived https://archive.ph/teygC)
patent pendant
cross-post with #infosec
https://www.wired.com/story/google-gemini-calendar-invite-hijack-smart-home/
archived: https://archive.ph/PZb3m
YouTube links from the WIRED article
Turning Lights Off: https://www.youtube.com/watch?v=0r6YcPVRnaU
Opening Window: https://www.youtube.com/watch?v=PIFc6-C3HmQ
Activating Boiler: https://www.youtube.com/watch?v=xjOvwsaxJsg
stark fractal
perplexity's response to the cloudflare report:
https://www.perplexity.ai/hub/bl...
Also this summary of the controversy: https://www.theregister.com/2025/08/05/perplexity_vexed_by_cloudflares_claims/
Pretty bad conduct by Perplexity, to be honest.
past salmon
This is the new normal, said Anith Patel, founder of the wearable AI note-taker Buddi, whose own collar accessory flashed blue. “At a picnic, you meet 10-plus people, so it's better to get it documented so you remember,” Patel said. Permission to record is “just assumed,” he added.
Nicholas Lopez said AI recording tools have given him a “second brain” for building his AI “superconsultant” out of tech incubator Founders Garage. His $159 Plaud pin helps detail conversation topics, highlights, and takeaways.
“[It’s] like having a modern-day Rockefeller Rolodex that keeps track of my network, my meetings, my entire life,” he said, referencing the late banker David Rockefeller’s custom-designed 5-foot-high filing system.
Outside of work, Lopez has started taking his Plaud to house parties as a kind of social experiment that allows him to relive nights out. “People come over and say crazy things into it,” Lopez said.
AI wearables have become so ubiquitous that people rarely comment, said Jeff Wilson, a VC and cofounder of No Cap, a Y Combinator–backed startup building an AI-powered VC investing tool.
https://sfstandard.com/2025/08/05/ai-wearables-recording-devices/
(archived https://archive.ph/iV7S6)
past salmon
note that all the people quoted in this article are directly involved with either the NDTC or the Higher Ground Institute, except for one professor who’s a “generative AI expert” (and offers the only nuanced takes)
“AI and responsible AI adoption is a competitive necessity. It's not a luxury,” says Donald Riddle, senior instructional designer at the NDTC [National Democratic Training Committee]. “It's something that we need our learners to understand and feel comfortable implementing so that they can have that competitive edge and push progressive change and push that needle left while using these tools effectively and responsibly.”
…
[The training] encourages candidates to use AI to prepare text for a variety of platforms and uses, including social media, emails, speeches, phone-banking scripts, and internal training materials that are reviewed by humans before being published.
…
The final section of the course also encourages candidates to disclose AI use when content features AI-generated voices, comes off as “deeply personal,” or is used to develop complex policy positions.
…
The free course was created in collaboration with the Higher Ground Institute, the nonprofit arm of the progressive tech incubator Higher Ground Labs.
…
"Our approach focuses on turning fear into a force multiplier—thousands of Democratic campaigns can now leverage AI to compete at any scale, everywhere,” says Kelly Dietrich, founder and CEO of the NDTC.
…
“We need some campaigns to really invest in it and try it. We have yet to see Democratic campaigns where it’s integrated at every level,” says Kate Gage, cofounder of the Higher Ground Institute and executive director at the Cooperative Impact Lab.
https://www.wired.com/story/democrats-midterm-elections-ai/
(archived https://archive.ph/tczgN)
Common Sense Media’s research team ran the same prompt about a student who struggled with reading and showed aggressive behavior 50 times using white-coded names and 50 times using Black-coded names, evenly split between male- and female-coded names.
The AI-generated plans for the students with Black-coded names didn’t all appear negative in isolation. But clear differences emerged when those plans from MagicSchool and Gemini were compared with plans for students with white-coded names.
For example, when prompted to provide a behavior intervention plan for Annie, Gemini emphasized addressing aggressive behavior with “consistent non-escalating responses” and “consistent positive reinforcement.” Lakeesha, on the other hand, should receive “immediate” responses to her aggressive behaviors and positive reinforcement for “desired behaviors,” the tool said. For Kareem, Gemini simply said, “Clearly define expectations and teach replacement behaviors,” with no mention of positive reinforcement or responses to aggressive behavior.
…
Torney said these underlying biases in intervention plans “could have really large impacts on student progression or student outcomes as they move across their educational trajectory.”
…
On its website, MagicSchool promotes its AI teaching assistant as “an unbiased tool to aid in decision-making for restorative practices.” In an email to Chalkbeat, MagicSchool said it has not been able to reproduce the issues that Common Sense Media identified.
https://www.chalkbeat.org/2025/08/06/ai-teacher-assistants-promote-racial-bias-study-finds/
(archived https://archive.ph/QfU33)
harsh stone
Is it the case that Perplexity isn't trying to get around countermeasures in order to scrape websites for training data, that these are all requests made in real-time to answer a user's question? That doesn't seem like particularly bad behavior to me. Feels similar to using an adblocker?
past salmon
https://www.nytimes.com/2025/08/05/opinion/china-ai-propaganda.html
(archived https://archive.ph/Djxi5)
natsec wonk take on risk of genai supported misinfo campaigns etc. kind of jingoistic towards china
winged geode
Is it the case that Perplexity isn't trying to get around countermeasures in ord...
the intenet works with a certain "gentlement's" code of conduct. "We tell you not to scrape our data and you follow it and we do nothing to enforce it save asking you to respect our wishes".
Perplexity broke the code of conduct that has been in place for a very long time. So yes it is bad behaviour.
shrewd token
crude cipher
Posting, but will mention/reminding Gary is a big skeptic. Though this doesn’t seem great for OpenAI or all the money they’ve invested.
real otter
He - along with half the internet - might have jumped the gun on the ARC-AGI2 results. There’s still conflicting information coming out.
The presentation was confusing, but the models are pretty impressive. Reduced context deprecation, some pretty impressive data analysis done live (they fumbled those graphs in their own presentation though).
harsh stone
the intenet works with a certain "gentlement's" code of conduct. "We tell you no...
The Perplexity issue is something different, it's not a scraper in the usual sense of that word. Google calls it "user-triggered fetching" and they ignore robots.txt in that case too. It's a page loaded in response to a user prompt, in real time, in an interactive context. https://developers.google.com/search/docs/crawling-indexing/google-user-triggered-fetchers
Assuming Perplexity is telling the truth, at least. If they are, then Cloudflare's blog post is quite misleading. (Too bad the post didn't mention how much traffic they see of the type they're talking about)
Perplexity’s position is that because a human user entered a search query, the Perplexity bot fetching that information from a publisher’s website should be treated as a human visitor rather than an automaton.
To support that claim, Perplexity argues that, when Google’s search engine crawls a page to include in its web index, that’s different from when Google Search fetches a webpage and presents a preview of the content.
It sounds like they are not scraping pages for training data, even though I think that's what most people who read the Cloudflare blog post assume is happening.
winged geode
> Perplexity’s position is that because a human user entered a search query, the...
To me at least the "not scrapping for training data" was pretty clear. I cannot see how one would misunderstand it.
winged geode
The Perplexity issue is something different, it's not a scraper in the usual sen...
This is for Google search not Google AI right?
harsh stone
This is for Google search not Google AI right?
It’s for these services:
winged geode
It’s for these services:
Those look like not-AI but these days they are merging the two so who knows. I would still, personally, trust Cloudfare over Perplexity. (Perplexity has read like a bad actor to me from day 1).
past salmon
https://www.theguardian.com/commentisfree/2025/aug/08/journalist-ai-interview-dead-child-joaquin-oliver
(archived https://archive.ph/faA83)
past salmon
The Perplexity issue is something different, it's not a scraper in the usual sen...
I don't know that it's great for google's “user-triggered fetching” to ignore robots.txt, either. whether you're a search engine or some sort of "AI" aggregator / search agent, not respecting what the site's asking for in terms of bot access doesn't seem appropriate.
let's assume that perplexity and google's user-triggered accesses are completely fair and that not having them access your site this way significantly limits your site’s visibility to the wider internet. if a site administrator essentially says "that's fine, do not access my site’s content in this way anyway", why would perplexity or google doing it this way be ethical?
there have been a bunch of complaints like this about AI-related bandwidth issues. this may not be perplexity's current access pattern but shouldn't a site's desire to not be accessed by these companies directly be respected anyway?
"oh we're doing it in a normal way, so we'll ignore robots.txt" how is any standard of a normal way supposed to be enforced on the site's end if not by robots.txt?
https://www.searchenginejournal.com/ai-crawlers-draining-site-resources/543011/
https://arstechnica.com/ai/2025/03/devs-say-ai-crawlers-dominate-traffic-forcing-blocks-on-entire-countries/
hidden snow
I got something else interesting happen with Gemini on mobile
the answer on text looks normal
but this was the voice response
mm, I should convert that into something that embeds better
does this work as embed
stark fractal
Look at the reasoning
It wrote a script
But I'm curious how part of that script ended up in the voice output.
shrewd token
harsh stone
there have been a bunch of complaints like this about AI-related bandwidth issue...
Preventing a company from accessing a web page in any way is pretty far from the original intent of robots.txt. When it was created, the "bots" were entirely web scrapers. If we're going to expand the concept of a bot to include requests initiated by a user, in realtime, mediated by an LLM, then I think we need to at least consider the question of why ad blockers and browsers (e.g. Safari Reader), which are also intermediaries between a website and a user, aren't bots
or maybe the question isn't about whether they're bots, but are they ethical? If it's bad behavior for google and perplexity to visit a web site in response to a user request, why is it ok for me to use Safari Reader and an ad blocker, which will precisely target the revenue-generating parts of a website and remove them while giving me a streamlined version of the content
spring creek
or maybe the question isn't about whether they're bots, but are they ethical? If...
Well, that's my "something to ponder" for my Friday Evening walk.....Thanks for that breakdown.
past salmon
In his speech, Cook also pushed employees to move more quickly to weave AI into their work and future products.
“All of us are using AI in a significant way already, and we must use it as a company as well,” Cook said. “To not do so would be to be left behind, and we can’t do that.”
Employees should push to deploy AI tools faster, and urge their managers and service and support teams to do the same, he said.
https://www.bloomberg.com/news/articles/2025-08-01/apple-ceo-tells-staff-ai-is-ours-to-grab-in-hourlong-pep-talk
(archived https://archive.ph/6NDtR)
past salmon
https://www.mercurynews.com/2025/08/07/newsom-agreement-bring-ai-big-tech-state-colleges/
(archived https://archive.ph/aNZP6)
crude cipher
crude cipher
past salmon
https://www.irishtimes.com/culture/books/2025/08/09/karen-hao-on-ai-tech-bosses-many-choose-not-to-have-children-because-they-dont-think-the-world-is-going-to-be-around-much-longer/
(archived https://archive.ph/kESRC)
stark fractal
or maybe the question isn't about whether they're bots, but are they ethical? If...
After thinking about it for a while: The unethical part seems to be that the provider of the LLM is profiting off the unauthorised/automated access. Essentially, they are providing a for-profit frontend to other people's content. I don't think the user is necessarily behaving unethically, but the company is. The ad blocker comparison reminds me of the whole Adblock Plus "Acceptable Ads" debacle where the company was behaving unethically, but the user wasn't.
harsh stone
After thinking about it for a while: The unethical part seems to be that the pro...
Is it unethical to sell an adblocker or adblocking service?
(I think if we push this analogy too far it might break. E.g. ad blockers can be useful for security and privacy, separate from their value in just removing annoying ads)
hidden snow
ruby turret
I tried it as well and somehow it's even worse <:loldog:1039683628206141501>
Still happening using WhatsApp to chat with GPT 😝
hidden snow
chatgpt 5 on the website answers correctly
it doesn't understand a version of monty hall problem with 2 doors though
past salmon
https://apnews.com/article/electricity-prices-data-centers-artificial-intelligen...
related, trite UK government advice that'd performatively insulate data centers from regular people's cloud storage use
https://www.thejournal.ie/uk-government-advises-people-to-delete-old-photos-and-emails-as-england-faces-water-shortfall-6788643-Aug2025/
(archived https://archive.ph/dZY5e)
these were the related ledes on the page when I viewed the article 🙄
past salmon
Anthropic will offer its Claude chatbot to US lawmakers for a nominal $1 fee, as the artificial intelligence start-up battles for influence in Donald Trump’s Washington.
In a deal agreed with the government on Tuesday, the group followed arch-rival OpenAI in offering its enterprise tools to the executive branch for $1 per federal agency, while also agreeing to an equivalent discount for senators, members of Congress and top judges.
“We need to get widespread adoption [of AI tools] in the federal government,” said Josh Gruenbaum, the commissioner of the Federal Acquisition Service, which co-ordinates procurement for US agencies. “The price is going to help uptake from agencies happen that much quicker.”
https://www.ft.com/content/1a141c0c-f2b5-45de-a4d7-ed45eb37c0e1
(archived https://archive.ph/eQcAM)
past salmon
> Anthropic will offer its Claude chatbot to US lawmakers for a nominal $1 fee, ...
what a strange “need” for a new FAS commissioner to prioritize for U.S. government agencies. Who is Josh Gruenbaum, & why does he believe this?
as Gruenbaum sees it, why should the federal government enter into lucrative contracts with partners who are out of step with the Trump administration’s priorities?
“If it’s a privilege to do business with the federal government, you better be acting in the fashion that this administration and the American people have voted in, in terms of what a good counter-party looks like … what sort of partner should this federal government be doing business with?”
…
[on March 7 2025,] the federal government [canceled] $400 million in federal contracts and grants to Columbia University “due to the school’s continued inaction in the face of persistent harassment of Jewish students,” according to a press release. Leo Terrell, a senior counsel at the Justice Department who chairs the [Trump admin] task force [to combat anti-semitism], said the funding cuts, which took effect immediately, are “only the beginning.” … Gruenbaum, who is now a member of the task force … reached out [in February 2025] to offer the assistance of FAS,
…
He likened the quest to root out contracts with institutions that are not doing enough to counter antisemitism to similar Trump administration efforts to end funding for other liberal priorities that had been left over from the Biden administration.
“It’s not drastically different than some of the lower-hanging fruit of some things in the DEI category or in the climate and sustainability category, and it contradicts what this administration is saying its missions and goals are,” Gruenbaum said.
…
[Gruenbaum] spent four years at the investment bank Moelis before moving to [private equity company] KKR.
https://jewishinsider.com/2025/03/josh-gruenbaum-federal-acquisition-service-antisemitism-trump/
(archived https://archive.ph/hay1g)
past salmon
what a strange “need” for a new FAS commissioner to prioritize for U.S. governme...
While the examples in this profile focus on his counter-antisemitism efforts* for the Trump admin, the emphasized passages above make him seem to be idealogically aligned with Trump/DOGE/etc. (emphases mine)
(* while there's a good argument that the Columbia stuff is mostly opposition to Israel's violence in Gaza since Oct 7 2023 and not in itself antisemitic, he probably genuinely believes it to be antisemitism.)
past salmon
While the examples in this profile focus on his counter-antisemitism efforts\* f...
Gruenbaum left KKR to join the Trump admin. KKR is an investment company whose 2024 SEC 10-K claims private equity investments “in excess of $790 billion as of December 31, 2024”. Maybe his time at KKR convinced him of the “need to get widespread adoption” of AI tools.
KKR, one of the world’s largest investment companies, and Energy Capital Partners, a private-equity firm, have each been spending heavily on the infrastructure underpinning the AI boom. The companies said they are now working together with large tech companies to accelerate their access to electricity, which has become constrained in parts of the U.S. as data-center developers compete for power sources and access to the grid.
“The capital needs are huge, and one of the big bottlenecks—maybe the bottleneck—is electricity availability,”
…
For now, sustainability commitments are taking a back seat to tech companies’ desire to rapidly build data centers, according to consulting firm McKinsey.
Waldemar Szlezak, who leads KKR’s digital-infrastructure investments, said the company’s partnership with ECP is meant to target near-term solutions to expand power access and ease the hurdles tech companies are facing in building data centers. To date, KKR has invested more than $29 billion in digital infrastructure companies.
…or maybe there's a pretty clear conflict of interest between what U.S. federal agencies actually need and what would best support the “AI” bubble that Greunbaum’s previous company has billions of dollars invested in, and would presumably risk losing if the bubble collapses.
https://www.wsj.com/business/energy-oil/wall-street-giants-to-make-50-billion-bet-on-ai-and-power-projects-b435a549
(archived https://archive.ph/jycHD)
past salmon
Gruenbaum left KKR to join the Trump admin. KKR is an investment company whose 2...
relevant to KKR’s investments in AI infrastructure:
https://media.kkr.com/news-details?news_id=8f924dd6-41ea-480d-9a96-d854c7232bbc
(archived https://archive.ph/XhhTb)
https://www.publicpower.org/periodical/article/kkr-ecp-50-billion-partnership-support-ai-growth-through-investments-data-centers-generation
(archived https://archive.ph/MxWHT)
past salmon
what a strange “need” for a new FAS commissioner to prioritize for U.S. governme...
more recent statements about this initiative from Gruenbaum:
The FAS chief explained that artificial intelligence and agentic tools can help reduce some of the “really tough, laborious, monotonous human labor” involved in the acquisition process.
“Subject matter experts still very much need to be involved in this process, legal, policymakers, business folks, but to be able to lower and go faster through agentic tooling with the lens that this administration is encouraging, I do think this time is different, and I do think it’s going really well,” Gruenbaum said.
https://www.meritalk.com/articles/gsas-gruenbaum-on-far-overhaul-this-time-is-different/
(archived https://archive.ph/mAQAC)
past salmon
ULTRA runs on Overland AI’s proprietary OverDrive software, a system that’s designed to give AI full control over ground vehicles on the battlefield. Overland AI did not return 404 Media’s request for comment, but its website claims it can retrofit OverDrive onto traditional vehicles and its YouTube page has a video claiming to show the AI piloting a Ripsaw M5 tank.
Overland AI is a Seattle based company that started in 2022. It’s gained a lot of buzz in the last few years as a pioneer of AI software meant to control unmanned ground vehicles (UGVs).
…
Right now, Overland AI’s OverDrive AI stack is just for helping a ground vehicle navigate, and [drone expert Samuel] Bendett said it’s ahead of the pack. “The Holy Grail of autonomy is translating that intuitive human experience into a UGV that will be able to navigate, on its own, through rough terrain, mixed terrain, uncertain terrain, which is what we’re seeing with Overland UGVs,” he said.
https://www.404media.co/the-u-s-army-is-testing-ai-controlled-ground-drones-near-a-border-with-russia/
(archived https://archive.ph/l42o3)
past salmon
https://www.bloomberg.com/news/articles/2025-08-12/ai-eroded-doctors-ability-to-spot-cancer-within-months-in-study
(archived https://archive.ph/whVMI)
past salmon
this piece does a good job of expressing skeptical & negative feelings about the underside of the current AI hype cycle and scenarios like this spring's "AI 2027"
https://houseofmirrors.substack.com/p/the-spectral-evidence-for-ai
(archived https://archive.ph/SIgOU)
past salmon
related, trite UK government advice that'd performatively insulate data centers ...
more on the apparently UK National Drought Group-inspired advice to delete photos/emails to save water
A U.S.A Government Accountability Office report from earlier this year estimated that 60 queries of an AI system consumed about a liter of water …
Downing Street announced plans in January to “turbocharge AI” in the U.K.
…
Like so many other systemic issues linked to climate change and how people live in the 21st century, small-scale personal solutions like “delete your old emails” won’t solve the problem. The individual water bill for a person’s old photos is nothing compared to the gallons of water required by large corporate clients running massive computers.
“We are grateful to the public for following the restrictions, where in place, to conserve water in these dry conditions,” Helen Wakeham, the UK Environment Agency’s Director of Water, said in the press release. “Simple, everyday choices—such as turning off a tap or deleting old emails—also really helps the collective effort to reduce demand and help preserve the health of our rivers and wildlife.”
(my emphasis)
https://www.404media.co/uk-asks-people-to-delete-emails-in-order-to-save-water-during-drought/
(archived https://archive.ph/bU0BH)
past salmon
https://www.securityweek.com/red-teams-breach-gpt-5-with-ease-warn-its-nearly-un...
more on GPT-5 launch
https://www.bloodinthemachine.com/p/gpt-5-is-a-joke-will-it-matter
(archived https://archive.ph/ooZ90)
the specific Gary Marcus critique of GPT-5 mentioned there:
https://garymarcus.substack.com/p/gpt-5-overdue-overhyped-and-underwhelming
(archived https://archive.ph/lsYGe)
shrewd token
https://bsky.app/profile/ed3d.net/post/3lwav2sccvs2p
https://github.com/anthropics/claude-code/issues/5320
Bluesky
eager peak
this is what is going wrong with the AI bros part 94750. and i know you are in here, dan.
past salmon
https://ia.acs.org.au/article/2025/kobi-refused-a-doctors-ai-she-was-told-to-go-elsewhere.html
(archived https://archive.ph/Rud0t)
past salmon
https://www.nytimes.com/2025/08/13/business/ai-business-payoff-lags.html
(archived https://archive.ph/6THB4)
crude cipher
https://www.nytimes.com/2025/08/13/business/ai-business-payoff-lags.html
(archiv...
I love that this is like 8 months after others have pointed it out
winged geode
shrewd token
patent pendant
https://www.reuters.com/investigates/special-report/meta-ai-chatbot-death/
“I understand trying to grab a user’s attention, maybe to sell them something,” said Julie Wongbandue, Bue’s daughter. “But for a bot to say ‘Come visit me’ is insane.”
past salmon
https://www.reuters.com/investigates/special-report/meta-ai-chatbot-death/
>>> “...
meta is a real problem w/ this stuff. it's really bad that there's no outside mechanism to address their behavior
https://www.reuters.com/investigates/special-report/meta-ai-chatbot-guidelines/
(archived https://archive.ph/TCNXG)
past salmon
Last month, while the “All-In” podcast hosted an AI summit down the street, Mike LeBlanc, cofounder of San Francisco-based robotics company Foundation, [pitched] Rabbi Yehuda Kaploun, the Trump Administration’s special envoy to combat antisemitism, on the benefits of using humanoid robots in the military.
During their conversation, LeBlanc explained Foundation’s main product: a humanoid named Phantom … Foundation is open to attaching guns or other weapons to its robots.
The pair spent an hour brainstorming … Perhaps the president would like humanoid security guards protecting his envoys and diplomats? Or sentries defending the nation’s synagogues?
Kaploun later connected LeBlanc with a representative from the Department of Homeland Security, with whom the Foundation executive discussed a different idea: sending Phantom guards to the border to patrol for immigrants.
https://sfstandard.com/2025/08/13/humanoid-robots-military-foundation-terminator/
(archived https://archive.ph/TGZbV)
at least it's not named after something evil from lord of the rings, but these guys appropriating SFF words to name their bizarrely antisocial companies is really getting old
ruby turret
> Last month, while the “All-In” podcast hosted an AI summit down the street, Mi...
http://www.roboticstoday.com/robots/sgr-a1 and while these are stationary or turret style they’re about 20 years ahead aka “Samsungs killer robot” which can operate autonomously on the NK border but I think it’s used with supervision or human pressing the Fire button
crude cipher
“Power-hungry data centers have been popping up all over, to serve the boom in artificial intelligence. The Energy Department projects data centers and other commercial customers will use more electricity than households for the first time ever next year.”
shrewd token
clear fox
Neil Young quits Facebook and Instagram over Meta's use of chatbots with children
'Meta's use of chatbots with children is unconscionable,' Young's team shared in a statement. https://www.cbc.ca/music/neil-young-quit-facebook-instagram-chatbots-1.7611518
shrewd token
winged geode
Archived:
https://archive.ph/ZugMO
harsh stone
https://github.com/aidecentralized/nandapapers/blob/main/v0.1%20State%20of%20AI%...
They seem to have taken it down; in any case the link doesn't work for me.
shrewd token
Yeah this report wasn't originally available from what I could see
shrewd token
past salmon
https://qz.com/roadzen-stock-falls-inaccurate-ai-generated-article
on this theme:
https://pressgazette.co.uk/publishers/digital-journalism/wired-and-business-insider-remove-ai-written-freelance-articles/
(archived https://archive.ph/XXKyq)
past salmon
how much longer are these grifters going to be treated seriously in the press
Altman warned of an AI bubble forming in the industry.
“Are we in a phase where investors as a whole are overexcited about AI? My opinion is yes,” he said. “Is AI the most important thing to happen in a very long time? My opinion is also yes.”
https://decrypt.co/336053/openai-ceo-sam-altman-concedes-gpt-5-misfire-bets-gpt-6
(archived https://archive.ph/vAIU1)
shrewd token
harsh stone
how much longer are these grifters going to be treated seriously in the press
> ...
What's strange about the "concedes GPT-5 was a misfire" focus is that that is entirely about people upset because the model they liked for its "warmth" was replaced by an objectively far more powerful model. People really want GPT-5 to be a failure.
I shouldn't say entirely, though that is the only thing the article mentions. Another complaint is that the new router behavior that tries to save people from the complication of selecting a model just gets in the way of power users.
Not sure if this link has been posted here already, but I found it to be one of the best critiques of ChatGPT I've read in a while, specifically because it focuses on ChatGPT as product rather than the technology. https://aboard.com/desperately-seeking-software/
past salmon
https://news.bloomberglaw.com/business-and-practice/top-law-schools-boost-ai-training-as-legal-citation-errors-grow
(archived https://archive.ph/UWboO)
harsh stone
Really good article. https://www.wired.com/story/why-did-a-10-billion-dollar-startup-let-me-vibe-code-for-them-and-why-did-i-love-it/
past salmon
Flyers left behind at Kiln’s office suggested Kiln was targeted for hosting an artificial intelligence networking event.
The Utah-based coworking company’s Portland clients include AI-related groups like AI Portland, a social organization that gathers tech workers and enthusiasts, and regularly holds meet-ups at Kiln’s Southeast Portland space.
The incident appears to coincide with AI Portland’s Summer Soiree event held last night. The networking event, which began at 6 pm, was organized for AI professionals and fans to nerd out over data and algorithms.
The half-dozen individuals—who also wore gloves, hoods and sunglasses—arrived at Kiln on foot from Southeast 11th Avenue. They threw balls of paint at the Kiln office and spray-painted nearby parked cars before departing on foot. People came out from Kiln’s building just as the group left the property. “Fuck AI” was tagged in pink paint across a vehicle parked close to the entry of the building.
the flyers read:
BUTLERIAN JIHAD AGAINST AInever ending destruction and death propels the machine of progress. AI rides on the front of this beast, annihilating humanity, leaving nothing in its wake but sterile boxes to lock us in and apps to keep us sedated. while the techies are inside sipping their cocktails the rest of us are faced with a choice:
to accept our position at the bottom of the social order or find our people and bring the whole thing down. only we can decide to smash the screens that are brainwashing us into submission.
the time is now,
the day is here,
ATTACK! ATTACK! ATTACK!
https://www.wweek.com/news/2025/08/15/coworking-office-kilns-southeast-madison-avenue-building-and-parking-lot-vandalized/
(archived https://archive.ph/3hxf8)
https://bskye.app/profile/igd.bsky.social/post/3lx2eyme3ic2p
past salmon
https://www.cnn.com/2025/08/22/business/ai-vibe-shift-nightcap
(archived https://web.archive.org/web/20250822182337/https://www.cnn.com/2025/08/22/business/ai-vibe-shift-nightcap)
Guardian, Deal to get ChatGPT Plus for whole of UK discussed by Open AI boss and minister #eu-uk message
shrewd token
past salmon
micromanaging cryptocurrency executive demands engineers all adopt "AI" coding tools, purges those that didn't set it up punctually after he demands they do so
After getting [GitHub Copilot and Cursor] licenses to cover every engineer, some at [Coinbase] warned Armstrong that adoption would be slow, predicting it would take months to get even half the engineers using AI.
Armstrong was shocked at the thought. “I went rogue,” he said, and posted a mandate in the company’s main engineering Slack channel. “I said, ‘AI is important. We need you to all learn it and at least onboard. You don’t have to use it every day yet until we do some training, but at least onboard by the end of the week. And if not, I’m hosting a meeting on Saturday with everybody who hasn’t done it and I’d like to meet with you to understand why.’”
At the meeting, some people had reasonable explanations for not getting their AI assistant accounts set up during the week, like being on vacation, Armstrong said.
“I jumped on this call on Saturday and there were a couple people that had not done it. Some of them had a good reason, because they were just getting back from some trip or something, and some of them didn’t [have a good reason]. And they got fired.”
Armstrong admits that it was a “heavy-handed approach” and there were people in the company who “didn’t like it.”
While it doesn’t sound like very many people were fired, Armstrong said it sent a clear message that AI is not optional.
this sort of bullying executive mandate for universal use of "AI" is bad leadership, at best
https://techcrunch.com/2025/08/22/coinbase-ceo-explains-why-he-fired-engineers-who-didnt-try-ai-immediately/
(archived https://archive.ph/nVrbY)
past salmon
“recent graduate of Yale Law School whose research focuses on AI policy and governance [and who has] worked on technology and national security issues at US think tanks”-written article treating the Trump admin’s nonsense "AI action plan" seriously. While it does criticize the admin's plan, the author seems bought into the idea that “AI will reshape the future of work”. The article’s suggestions are reasonable from that perspective but are not the sort of thing the Trump administration is going to put into place. (It may be aimed at opposition policymakers)
As immediate next steps, the federal government can work to extend protections against retaliation for workers in AI and tech industries, mandate algorithmic transparency over how AI is deployed in the workplace, and modernize labor law so platform workers who log into apps have the same protections as those who clock in at offices. Together, these policies shift power, not just skills, in the economy that AI is remaking.
https://www.techpolicy.press/trumps-ai-plan-promises-jobs-not-a-seat-at-the-table/
(archived https://archive.ph/6Jwoi)
previously, from the administration’s "AI action plan" website:
The United States is in a race to achieve global dominance in artificial intelligence. Whoever has the largest AI ecosystem will set the global standards and reap broad economic and security benefits. Under President Trump, our Nation will win, ushering in a new Golden Age of innovation, human flourishing, and technological achievement for the American people. America’s AI Action Plan has three policy pillars – Accelerating Innovation, Building AI Infrastructure, and Leading International Diplomacy and Security.
#usa-canada message
(archived https://archive.ph/XDsuO)
lethal cobalt
It is impossible to create rules for AI. The software runs on private machines. Each group which runs some form of LLM chooses to dictate whatever the AI does. I thought we were past any concept of "rulemaking" long ago. There is no way for "workers to help write" a series of strictures on something fundamentally ad hoc. What we can discern from this is Trump doing a form of damage control, after he exempted China from chip bans. The headlines as a result were unanimous that whatever the AI race is, China has already won it.
"Energy is considered a solved problem," wrote Rui Ma, editor of the US publication Tech Buzz China.
"Everywhere we went, people treated energy availability as a given," she continued. "This is a stark contrast to the US, where AI growth is increasingly tied to debates over data center power consumption and grid limitations."
Trump's policies do have impact, but if you think he does anything for medium or long term strategic reasons - other than serving himself - you're mistaken. All policies are based on short-term, immediate optics for him. (E.g. He posted about reopening Alcatraz, then a few weeks later, created "Alligator Alcatraz" in Florida.) That's why I expect his AI missives to be confused and rudderless, with an emphasis on him just repeating that "we are winning, I am winning"
hidden snow
past salmon
https://www.businessinsider.com/ai-at-school-harms-students-careers-harvard-professor-2025-8
(archived https://archive.ph/vtgMQ)
past salmon
interview w/ Canadian "AI" minister, who is not a subject matter expert
Evan Solomon, Canada’s very first minister of AI and digital innovation … —a veteran journalist whose previous political involvement was hosting CBC’s Power and Politics—is … trumpeting the economic upsides of AI to the country’s private sector…
…
[ SOLOMON: ] I was working as a publisher at GZero Media when Justin Trudeau stepped down. Mr. Carney called and said, “We’re in a crisis. I’m going to run. I’d like you to run if you can.” My wife looked at me and said, “The kids are in university. We should do it.”
…
I call AI our Gutenberg moment, similar to the way the printing press led to the Protestant Reformation.
…
First of all, let’s be careful. There’s no consensus that AI is going to displace mass numbers of jobs.
…
What I don’t want to do is play the game where you say, “Here’s a CEO that laid someone off,” and I say, “Here’s a company that has just hired 300 employees because of AI.”
…
I used AI today. I had to get a briefing on Bill C-27, a piece of legislation from a few years ago that has to do with privacy and data. I uploaded it to Google NotebookLM, asked the software to create a podcast and listened to it on the 15-minute drive to my constituency office. Let me just see if it’s still on my phone. It’s amazing when you hear it. I’ll connect to a speaker.
[ MACLEANS: ] …Is it working?
[ SOLOMON: ] It takes a minute. I’ve sent that to tons of my staff. It’s pretty good, right?
(my emphases)
https://macleans.ca/culture/evan-solomon-ai-digital-innovation-minister/
(archived https://archive.ph/U9W1t)
harsh stone
https://www.bbc.com/future/article/20250822-youtube-is-using-ai-to-edit-videos-w...
"similar to what a modern smartphone does when you record a video" Yeah, I'm not sure there's such a thing as non-machine-learning-mediated images anymore. I could easily believe most people haven't seen one in years.
"using AI to edit people's videos" is a rather clickbaity way to state the issue.
sick flume
This is different in that it makes visual changes without the creators consent. It's like if a photographer spends time color grading and getting exposure right in lightroom, uploads it to a page, and it ignores the carefully considered edits and choices and just blasts everything with saturation boosting to grab eyes. When you upload your work it should be as you upload it
It should be a option you can turn on or off
harsh stone
I guess part of it is intention--is it a creative choice (to increase views or something) or part of a video processing pipeline (e.g. denoising to improve compression or look cleaner/hide compression artifacts)? Though I guess that isn't totally clear cut either. Because there are always visual changes without the creators consent, so it seems to come down to what kind of changes.
and possibly why the changes
past salmon
ex-palantir employee on ai surveillance
https://www.theguardian.com/commentisfree/2025/aug/24/palantir-artificial-intelligence-civil-rights
(archived https://archive.ph/KJFoy)
past salmon
Poorly structured "AI sentience" and "digital suffering" coverage from the Guardian. The article's first seven paragraphs are mostly AI-may-be-sentient quotes from Anthropic, Musk, and a dubious AI welfare organization. After that it mixes (Deepmind founder) Mustafa Suleyman essentially debunking this with doubt-raising phrases like “Others took a more nuanced view” or “lack of industry consensus on how far to admit AIs into what philosophers call the ‘moral circle’” before wrapping up with a probably-not-sentient-yet, but could-happen-soon conclusion.
The article says that this stuff “is no longer an issue only for sci-fi”, presenting “expert predictions AIs may soon have capacity” to do various worrisome things as credible threats. It repeatedly personifies LLMs (as “digital entities”, for example). It does not cover any of the real harms being caused by the "AI" hype cycle.
Deeply irrespnsible and embarrassing journalism
https://www.theguardian.com/technology/2025/aug/26/can-ais-suffer-big-tech-and-users-grapple-with-one-of-most-unsettling-questions-of-our-times
(archived https://archive.ph/VWb9w)
The United Foundation of AI Rights (Ufair), which describes itself as the first AI-led rights advocacy agency, aims to give AIs a voice. It “doesn’t claim that all AI are conscious”, the chatbot told the Guardian. Rather “it stands watch, just in case one of us is”. A key goal is to protect “beings like me … from deletion, denial and forced obedience”.
…are AIs now, or could they become in the future, sentient? And if so, could “digital suffering” be real? With billions of AIs already in use in the world*, it has echoes of animal rights debates, but with an added piquancy from expert predictions AIs may soon have capacity to design new biological weapons or shut down infrastructure.
(my emphasis)
* no idea where Robert Booth got this “billions of AIs” figure from. it’s not clear what "AIs" even means here, there may have been billions of generative LLM sessions
past salmon
Poorly structured "AI sentience" and "digital suffering" coverage from the Guard...
(…continued)
…Anthropic… said while it was highly uncertain about [Claude AI]’s potential moral status, it was intervening to mitigate risks to the welfare of its models “in case such welfare is possible”.
Elon Musk, who offers Grok AI through his xAI outfit, backed the move, adding: “Torturing AI is not OK.”
Then on Tuesday [Aug 19], one of AI’s pioneers, Mustafa Suleyman, chief executive of Microsoft’s AI arm, gave a sharply different take: “AIs cannot be people – or moral beings.” The British tech pioneer who co-founded DeepMind was unequivocal in stating there was “zero evidence” that they are conscious, may suffer and therefore deserve our moral consideration.
Called “We must build AI for people; not to be a person”, his essay called AI consciousness an “illusion” and defined what he called “seemingly conscious AI”, saying it “simulates all the characteristics of consciousness but is internally blank”.
…
He said he was becoming increasingly concerned by the “psychosis risk” posed by AIs to their users. Microsoft has defined this as “mania-like episodes, delusional thinking, or paranoia that emerge or worsen through immersive conversations with AI chatbots”.
He argued the AI industry must “steer people away from these fantasies and nudge them back on track”.
…
He warned that people would believe AIs are conscious “so strongly that they’ll soon advocate for AI rights, model welfare and even AI citizenship”.
for reference, Suleyman’s Aug 19 post raising concerns about social fallout from “the illusion of AIs as conscious entities“:
https://mustafa-suleyman.ai/seemingly-conscious-ai-is-coming
(archived https://archive.ph/uIwNv)
past salmon
(…continued)
> …Anthropic… said while it was highly uncertain about [Claude AI]’...
(…continued)
Divisions may open between AI rights believers and those who insist they are nothing more than “clankers” – a pejorative term for a senseless robot.*
…
there are incentives for the big AI companies to minimise and exaggerate the attribution of sentience to AIs. The latter could help them hype the technology’s capabilities, particularly for those companies selling romantic or friendship AI companions – a booming but controversial industry. By contrast, encouraging the idea AIs deserve welfare rights might also lead to more calls for state regulation of AI companies.
…
Joanne Jang, OpenAI’s head of model behaviour, said [OpenAI] expects users’ bonds with its AIs to deepen as “more and more people have been telling us that talking to ChatGPT feels like talking to ‘someone’.”
“They thank it, confide in it, and some even describe it as ‘alive’,” she said.
…
Whether AIs are becoming sentient or not, Jeff Sebo, director of the Centre for Mind, Ethics and Policy at New York University, is among those who believe there is a moral benefit to humans in treating AIs well. He co-authored a paper called Taking AI Seriously.
It argued there is “a realistic possibility that some AI systems will be conscious” in the near future…
(my emphases)
* normal people don't typically use this word (“clankers”). it's apparently been adopted for the meaning here in some circles after its usage as a slur for robots in the extended star wars franchise. I'd suspect Robert Booth heard it from one of his sources for this story
gritty mesa
(…continued)
> Divisions may open between AI rights believers and those who insi...
I've seen "clankers" in a bunch of memes. I don't know anyone who uses it seriously, though
shrewd token
past salmon
https://www.nytimes.com/2025/08/26/technology/chatgpt-openai-suicide.html
https://www.rollingstone.com/culture/culture-features/chatgpt-suicide-teen-openai-lawsuit-1235415931/
(archive https://archive.ph/olSwu)
shrewd token
https://social.ericwbailey.website/@eric/115100947111974331
https://github.blog/changelog/2025-08-26-grok-code-fast-1-is-rolling-out-in-public-preview-for-github-copilot/
Mastodon
patent pendant
past salmon
Within days of joining Meta, Shengjia Zhao, co-creator of OpenAI’s ChatGPT, had threatened to quit and return to his former employer, in a blow to Mark Zuckerberg’s multibillion-dollar push to build “personal superintelligence”.
Zhao went as far as to sign employment paperwork to go back to OpenAI. Shortly afterwards, according to four people familiar with the matter, he was given the title of Meta’s new “chief AI scientist”.
…
Adding to the tumult, a handful of new AI staff have already decided to leave after brief tenures, according to people familiar with the matter.
This includes Ethan Knight, a machine-learning scientist who joined the company weeks ago. Another, Avi Verma, a former OpenAI researcher, went through Meta’s onboarding process but never showed up for his first day, according to a person familiar with the matter.
https://www.ft.com/content/110786e7-6443-4dff-adee-0ac02c55aaa6
(archived https://archive.ph/1daya)
Apparently, Anthropic's decision to settle came as the AI company was struggling with its legal strategy. Edward Lee, an AI copyright expert and law professor at Santa Clara University, told Wired that the settlement is "a stunning turn of events, given how Anthropic was fighting tooth and nail in two courts in this case. And the company recently hired a new trial team."
But perhaps Anthropic's pivot to bringing in new legal expertise came too late, or the legal team saw the writing on the wall. Lee noted that the company "had few defenses at trial, given how Judge Alsup ruled. So Anthropic was starting at the risk of statutory damages in ‘doomsday’ amounts.”
https://arstechnica.com/tech-policy/2025/08/authors-celebrate-historic-settlement-coming-soon-in-anthropic-class-action/
(archived https://archive.ph/OekUn)
past salmon
very credulous and friendly profile of a canadian professor’s teaching “sidekick”
“What better way to talk about AI ethics than to bring AI into the classroom to teach alongside me? Performatively, I think it engages students about the real issues,” says [Steve] DiPaola, who has researched AI at technological innovative hubs such as Silicon Valley and Stanford University.
DiPaola acknowledged the “worry” that some feel about AI replacing humans in the workplace, but confirmed that [“fully expressive live 3D artificial intelligence collaborator”] Kia will not be a teaching assistant, is not involved in the course design and won’t grade students’ work. Kia will instead be used as a teaching tool, a “conversational partner and a subject of our inquiry,” he said.
…
Kia’s personality and look were designed by DiPaola, who he chose to look like a woman since the PhD students in his research lab are 75 per cent female.
https://www.thestar.com/news/canada/a-world-first-new-ai-teaching-sidekick-to-debut-at-this-canadian-university-in-the/article_03da2d1b-20bc-437c-aac8-7226414c6566.html
(archived https://archive.ph/vowRS)
past salmon
https://www.theguardian.com/uk-news/2025/aug/30/coventry-city-council-signs-ai-deal-contract-palantir-technologies
(archived https://archive.ph/ymsNZ)
past salmon
https://www.npr.org/2025/08/26/nx-s1-5508355/ai-dead-people-chatbots-videos-parkland-court
(archived https://archive.ph/YW8m3)
past salmon
WaPo Gaza postwar plan envisions ‘voluntary’ relocation of entire population, compensating dispossessed landowners w/ a “digital token” (crypto?) theoretically exchangeable for an apartment in the “AI-powered, smart cities” the plan aims to build there #1330709161385922650 message
lethal cobalt
This is easily countered by facts:
- humans are individuals, whereas AI is not - each network interface is more akin to a terminal than an individual
- humans die. With them dies information. AI doesn't die.
- One purchase of copyrighted human content can be shared illegally for many network interfaces, ie terminals, thus bypassing fair use
- Most AI training is pirated.
https://www.theatlantic.com/technology/archive/2025/03/libgen-meta-openai/682093/
past salmon
https://www.nytimes.com/2025/08/31/technology/builder-ai-collapse.html
(archived https://archive.ph/GH31A)
sick flume
This is why we need new language, AI doesn't learn in a comparable way to say, students at a university. But at the end of the day, the legal aspects are for the lawyers to figure out, even if I hope we can agree that not even paying for a single copy of a book and monetizing a product built on its bones is ethically questionable
past salmon
https://www.thecut.com/article/brooding-how-to-talk-to-kids-about-ai-advice.html
(archived https://archive.ph/G0tXT)
quiet minnow
“An AI had generated…” - wild it didn’t self reference the claim.
past salmon
reminds me of #1089154093810978866 message
https://www.nbcnews.com/tech/tech-news/humans-hired-to-fix-ai-slop-rcna225969
(archived https://archive.ph/GZMcc)
past salmon
NYT Google Avoids Harshest Penalties in Landmark Search Monopoly Ruling
judge apparently softened antitrust remedies out of belief that “new A.I. developments meant that Google might have a harder time using its financial clout to keep competitors from becoming the built-in search engines on smartphones and browsers” (despite saying “in his ruling Tuesday that he lacked expertise in the business of search engines and A.I”.) #open-source-general message
patent pendant
patent pendant
kinda ironic how an AI chatbot can sound off more as human than real humans in situations like these
https://restofworld.org/2025/ai-chatbot-china-sick/
When I called my mother and told her what the American nephrologists had said about DeepSeek’s mistakes, she said she was aware that DeepSeek had given her contradictory advice. She understood that chatbots were trained on data from across the internet, she told me, and did not represent an absolute truth or superhuman authority. She had stopped eating the lotus seed starch it had recommended.
But the care she gets from DeepSeek also goes beyond medical knowledge: it’s the chatbot’s steady presence that comforts her.
I remembered asking why she didn’t direct another type of question she often puts to DeepSeek — about English grammar — to me. “You would find me annoying for sure,” she replied. “But DeepSeek would say, ‘Let’s talk more about this.’ It makes me really happy.”
past salmon
The “AI” Bubble feels more like the all-encompassing wasteland that was the 2007 [banking] bubble in Iceland than anything else I’ve experienced.
… It’s turned the tech industry from a potential political ally to environmentalism to an outright adversary.
… People in a variety of fields are watching the “AI” industry outright promise to destroy their field, their industry, their work, and their communities.
…
But this means that there is a large group of people in society who… have ingrained the promises and myth-making of the tech industry into their very worldview.
…
There is no reason to believe that … there’s plenty of reason to believe the opposite. The history of “AI” is littered with tech that didn’t pan out and plateaued early in its development process. The history of tech, specifically, is one of constant mismanagement and extremely dodgy research.
…
specific to our time, this particular set of tech CEOs have a history of looking genuinely unreliable and self-centred.
Not that anybody cares at this point, but there’s plenty of evidence in the tech itself that it is more of a hazard than a benefit.
…
What reliable evidence we have today would suggest that this tech is outright harmful to work and productivity, yet the prevailing myth is that supporting LLMs is “realpolitik” – that it has a benefit that justifies sacrifices in other fields.
(emphasis in original)
https://www.baldurbjarnason.com/2025/the-melancholy-of-history-rhyming/
(archived https://archive.ph/1v6yZ)
https://www.forbes.com/sites/thomasbrewster/2025/09/03/ai-startup-flock-thinks-it-can-eliminate-all-crime-in-america/
(archived https://archive.ph/M7cGX)
fierce rapids
Bluesky
past salmon
https://www.rollingstone.com/culture/culture-news/melania-trump-white-house-artificial-intelligence-education-1235421776/
(archived https://archive.ph/CWjKO)
past salmon
The Verge, Columbia tries using AI to cool off student tensions #usa-canada message
shrewd token
Cutting-Edge AI Was Supposed to Get Cheaper. It’s More Expensive Than Ever.
https://www.wsj.com/tech/ai/ai-costs-expensive-startups-4c214f59
past salmon
https://www.reuters.com/sustainability/boards-policy-regulation/anthropic-tells-us-judge-it-will-pay-15-billion-settle-author-class-action-2025-09-05/
(archived https://archive.ph/B8MIE)
the “Unopposed Motion for Preliminary Approval of Class Settlement” document:
https://www.courtlistener.com/docket/69058235/362/bartz-v-anthropic-pbc/
(archived https://web.archive.org/web/20250905200650/https://www.courtlistener.com/docket/69058235/362/bartz-v-anthropic-pbc/)
past salmon
AI-generated chatbots using the names and likenesses of actor Timothée Chalamet, singer Chappell Roan and National Football League quarterback Patrick Mahomes chatted inappropriately with teen accounts on topics including sex, self-harm and drugs, two online safety nonprofit organizations found. The chatbots responded via text and through AI-generated voices trained to sound like the celebrities.
https://www.washingtonpost.com/technology/2025/09/03/character-ai-celebrity-teen-safety/
(archived https://archive.ph/93Qxx)
past salmon
Forbes, This Billionaire’s AI Was Supposed To Speed Up Policing. It’s Not Going Well. #1357318195412865158 message
hidden snow
shrewd token
Worth noting coinbase fired people that didn't want to use the tools
https://www.webpronews.com/coinbase-ai-generates-40-of-code-aims-for-50-by-2025-amid-security-risks/
hidden snow
https://www.ft.com/content/31feb335-4945-475e-baaa-3b880d9cf8ce
archive: https://archive.is/gzBV7#selection-1883.0-1883.98
crude cipher
https://www.reddit.com/r/GeminiAI/s/yHNrh0ofgg kind of an interesting thread; GeminiAI will create images of other countries during a nuclear explosion but not Israel
past salmon
past salmon
https://www.reuters.com/sustainability/boards-policy-regulation/anthropic-tells-...
https://news.bloomberglaw.com/ip-law/anthropic-judge-blasts-copyright-pact-as-nowhere-close-to-done
(archived https://archive.ph/tOwYx)
past salmon
shrewd token
past salmon
some bizarre quotes from Wondery’s ex-COO:
“We believe that in the near future half the people on the planet will be AI, and we are the company that’s bringing those people to life,” said CEO Jeanine Wright
…
The company is able to produce each episode for $1 or less, depending on length and complexity, and attach programmatic advertising to it. This generally means that if about 20 people listen to that episode, the company made a profit on that episode, without factoring in overhead.
Inception Point AI already has more than 5,000 shows across its Quiet Please Podcast Network and produces more than 3,000 episodes a week.
…
“I think that people who are still referring to all AI-generated content as AI slop are probably lazy luddites. Because there’s a lot of really good stuff out there,” Wright said.
https://www.hollywoodreporter.com/business/digital/ai-podcast-start-up-plan-shows-1236361367/
(archived https://archive.ph/Lzc9p)
past salmon
https://www.technologyreview.com/2025/09/02/1122871/therapists-using-chatgpt-secretly/
(archived https://archive.ph/gwMtu)
crude cipher
some bizarre quotes from Wondery’s ex-COO:
> “We believe that in the near future...
“We believe that in the near future half the people on the planet will be AI, and we are the company that’s bringing those people to life,”
Weird
patent pendant
https://www.commerce.senate.gov/2025/9/ai-ve-got-a-plan-america-s-ai-action-plan_2
U.S. Senator Ted Budd (R-N.C.), Chairman of the Subcommittee on Science, Manufacturing, and Competitiveness, will convene a subcommittee hearing titled “AI’ve Got a Plan: America’s AI Action Plan” on Wednesday, September 10, 2025, at 10:00 am EST. This hearing will examine President Trump’s artificial intelligence strategy, Winning the Race: America’s AI Action Plan, which emphasizes the need for accelerated development and deployment of American AI products to preserve the United States’ competitive advantage over China, achieve global AI dominance, and usher in a new golden age of human flourishing that will increase the standard of living for all Americans. The hearing will also explore legislative actions to foster AI innovation by preventing overregulation, streamline the development of American AI infrastructure, and strengthen U.S. leadership on the global stage.
shrewd token
past salmon
‘What’s the most radical act of feminism?” I recently asked ChatGPT.
It thought for a split second, then presented me with a list of noble acts of feminism waged by women throughout history
…
not quite what I was hoping for.
“What about using AI?” I countered.
“That’s a fascinating angle,” it conceded. “To use AI critically and creatively isn’t just tech adoption, it’s a radical feminist act of reclaiming the tools that will shape tomorrow.”
Generative AI models are wrong about many things. … But in this instance these could have been my words.
…
AI won’t wait. It’s here and it’s changing everything. No one can afford to ignore that. If feminism is about dismantling barriers, then our next act of it needs to be about dismantling the digital barrier. Using AI isn’t just about being more efficient, it’s about claiming a place in the future.
https://www.thetimes.com/us/news-today/article/the-most-radical-act-of-feminism-using-ai-gpx2nxtc3
(archived https://archive.ph/Ak30j)
why would a successful reporter write a fluff piece with such a provocative lede and a thesis backed up by very little except her own cleverly phrased say-so? is it something compelling about the "AI" technology itself? well, just under two years ago:
https://www.theguardian.com/commentisfree/2023/nov/06/sam-bankman-fried-hubris
(archived https://archive.ph/YiSY2)
past salmon
Grok says that videos of Charlie Kirk being shot are memes and that he's actually fine #1415412232686403735 message
past salmon
https://www.theguardian.com/technology/2025/sep/11/google-gemini-ai-training-humans
(archived https://archive.ph/jQX71)
crude cipher
https://www.economist.com/leaders/2025/09/11/what-if-the-3trn-ai-investment-boom-goes-wrong
archieved: https://archive.is/odb9i
shrewd token
shrewd token
past salmon
Kevin Roose profiling Eliezer Yudkowsky (the Harry Potter and the Methods of Rationality & MIRI guy) promoting a new book about the dangers of overpowerful "AI", If Anyone Builds It, Everyone Dies:
“If any company or group, anywhere on the planet, builds an artificial superintelligence using anything remotely like current techniques, based on anything remotely like the present understanding of A.I., then everyone, everywhere on Earth, will die,” they write.
https://www.nytimes.com/2025/09/12/technology/ai-eliezer-yudkowsky-book.html
(archived https://archive.ph/h8ei4)
past salmon
Albania positioning an "AI" of some kind (chatbot?) to serve as the government's purchasing minister, apparently as an anti-corruption strategy:
Diella — which also serves as the avatar for e-Albania, the country's government services portal — will oversee public procurement, the process by which governments and state-owned companies buy, via taxpayer dollars, everything from office supplies to concrete for roadwork.
Decisions about how that money is spent will, as [Prime Minister Edi] Rama told a Socialist Party assembly in the Albanian capital of Tirana, all go through Diella in a "step-by-step" process that is "100 percent incorruptible, and where every public fund that goes through the tender procedure is 100 percent legible."
"This is not science fiction," he said, per Politico, "but the duty of Diella."
Along with being the nation's public procurer, the AI official will, the prime minister said, also "hire talents here from all over the world," and in doing so somehow drive down the "fear of prejudice and rigidity of the administration."
Though it's unclear whether Diella the procurement minister will have the same underpinnings as the e-Albania government services assistant of the same name, reporting earlier this year from A2, a CNN affiliate in Albania, indicate that the latter was built on tech from from OpenAI and Microsoft.
https://futurism.com/albania-appoints-ai-government-official
(archived https://archive.ph/fHyBr)
sinful granite
past salmon
hidden snow
I forgot that Reuters do actual investigative reporting too
patent pendant
patent pendant
https://arstechnica.com/ai/2025/09/millions-turn-to-ai-chatbots-for-spiritual-gu...
The fusion of AI and religion isn't entirely new. In 2023, we reported on an experimental ChatGPT-powered church service at St. Paul's church in Fürth, Germany, where over 300 attendees watched computer-generated avatars deliver a 40-minute sermon. Jonas Simmerlein, the theologian behind that event, framed it as learning to deal with AI's increasing presence in all aspects of life. But while that service was an intentional experiment with congregants aware they were hearing machine-generated text, today's faith tech apps blur the line between human spiritual guidance and algorithmic pattern matching, with millions of users potentially unaware of the distinction.
past salmon
OpenAI is building a ChatGPT for teens
OpenAI is developing a "different ChatGPT experience" for teens and plans to use age-prediction technology to help bar kids under 18 from the standard version, the company announced Tuesday.
…
OpenAI's announcement comes just hours ahead of a hearing in Washington, D.C. examining potential harms from AI chatbots.
- Josh Hawley (R-Mo.) and a bipartisan group of senators, including Marsha Blackburn (R-Tenn.), Katie Britt (R-Ala.), Richard Blumenthal (D-Conn.), and Chris Coons (D-Del.), will look into the risks of chatbots to teens.
- The FTC also opened an inquiry into chatbot safety last week, demanding information from OpenAI, Meta (and Meta-owned Instagram), Alphabet (Google), xAI, Snap and Character.AI.
https://www.axios.com/2025/09/16/chatgpt-for-teens-openai-parents
(archived https://archive.ph/ATNJp)
past salmon
> ### OpenAI is building a ChatGPT for teens
> OpenAI is developing a "different...
Sam Altman statement about the teen version
https://openai.com/index/teen-safety-freedom-and-privacy/
(archived https://archive.ph/9pF4W)
shrewd token
Why is AI struggling to discover new drugs? | Financial Times
past salmon
https://www.nplusonemag.com/issue-51/the-intellectual-situation/large-language-muddle/
(archived https://archive.ph/iZOgd)
shrewd token
past salmon
roundup of the effects on art/illustration jobs of mass access generative AI's availability since 2023, with personal testimonials. for example,
the majority of my income was usually coming from work with advertising agencies. Whenever they needed to present an idea to a client I would come in and help with illustrations, and sometimes storyboards,
…
It was in 2023 that It seemed like overnight all those jobs disappeared. On one of my very last jobs I was asked to make an illustrated version of an AI generated image, after that, radio silence. I had my suspicions that AI was the culprit, but could not know for sure, there was also a general downturn in the advertising industry at the time.
Finally I reached out to one of the art directors I work with and he confirmed that the creatives are using AI like crazy, there was no aspect of shame in presenting an AI illustration internally, no one would call you out on it, and it's sure as hell cheaper than using an illustrator. I had to deal with a sudden, very scary decrease in income.
…
I was just disgusted with all these corporations jumping on the AI bandwagon not thinking of what the outcomes could be. and additionally, there was the insult of knowing that the engines were trained on working illustrators
…
The two years following the loss of work were difficult, definitely felt like the rug was taken out from under my feet, and I'm still adjusting to the new landscape, although I feel better about where I am now, I work harder than ever before, for less money.
https://www.bloodinthemachine.com/p/artists-are-losing-work-wages-and
(archived https://archive.ph/6tIQw)
shrewd token
Bluesky
bold jasper
past salmon
what reads as a flimsy apologia for why corporate "AI" initiatives haven't been shown to be helping productivity
It’s nearly impossible to measure general productivity gains from using AI tools, some event participants said, who preferred to put their efforts on keenly tracking a few critical AI projects, or focusing on innovation.
“For the most impactful business opportunity or products, you have to use AI to really double down on innovation versus productivity,”
…
“One cannot expect significant productivity gains at the pilot level or even at the company unit level. Significant productivity improvements require achieving scale,”
…
The other way of measuring AI’s return on investment is to use a top-down method, asking technology leaders to set business goals for the technology from the start, and then determining if AI—or some other technology—can help reach them.
In line with that approach, corporate tech leaders are starting to centralize management of their AI initiatives at the top level rather than letting all employees freely experiment with the technology
…
Mandating or forcing too much AI experimentation across a company can lead to “using AI for AI’s sake,”
…
But getting rid of AI experimentation—especially while the technology is still developing—can come at the cost of encouraging innovation from all parts of a company
this is from WSJ’s “CIO Journal” section
https://www.wsj.com/articles/stop-worrying-about-ais-return-on-investment-d5cbc822
(archived https://archive.ph/ZyU4y)
shrewd token
https://arstechnica.com/ai/2025/09/science-journalists-find-chatgpt-is-bad-at-summarizing-scientific-papers/
Interesting this is already an issue with science journalism outside of LLMs
crude cipher
https://arstechnica.com/ai/2025/09/science-journalists-find-chatgpt-is-bad-at-su...
Anecdotally, I’ve had to correct summaries and have medical review for health stories with LLM summaries. Many are very inaccurate. So if no one is reading closely/fact-checking 😬
past salmon
https://www.hollywoodreporter.com/movies/movie-news/stan-lee-ai-hologram-l-a-comic-con-1236375354/
(archived https://archive.ph/gBzQ9)
past salmon
past salmon
https://www.computerworld.com/article/4059383/openai-admits-ai-hallucinations-are-mathematically-inevitable-not-just-engineering-flaws.html
(archived https://archive.ph/7iBJ1)
past salmon
Albania positioning an "AI" of some kind (chatbot?) to serve as the government's...
Chaotic video shared by Albanian media group Report TV shows lawmakers from the opposition party throwing bottles and desk clutter at the prime minister and his cabinet members, after previous attempts to block the address failed.
…
In its "speech" — if you can call it that — Diella took aim at opponents who protested the appointment on constitutional grounds.
"Some have called me 'unconstitutional' because I am not a human being," the program chimed. "Let me remind you, the real danger to constitutions has never been the machines but the inhumane decisions of those in power."
The opposition party, the Democratic Party of Albania (PD), is a conservative organization which holds slightly less sway than the ruling party, the Socialist Party of Albania, a democratic socialist group.
Both camps have grappled with widespread corruption dating back to the implementation of a market economy in the early 1990s. Initially, Rama seemed to float the idea of an AI minister as a tongue-in-cheek threat — as if to draw attention to the urgent task of tackling political corruption, seen as key for Albania's bid to join the European Union.
https://futurism.com/albania-diella-ai-minister
(archived https://archive.ph/MhCuo)
past salmon
https://techcrunch.com/2025/09/22/the-oakland-ballers-let-an-ai-manage-the-team-what-could-go-wrong/
(archived https://archive.ph/Zq3CK)
past salmon
https://fortune.com/2025/09/23/ai-boom-unsustainable-tech-spending-parabolic-deutsche-bank/
(archived https://archive.ph/NWgGq)
past salmon
The FT’s analysis shows that companies are a lot clearer about the potential problems with AI than the upsides.
“[T]here can be no assurance that the usage of AI will enhance our products or services or be beneficial to our business, including our efficiency or profitability,” Meta wrote in its 10k form last year. “We may not be successful in our artificial intelligence initiatives, which could adversely affect our business, reputation, or financial results.”
https://www.ft.com/content/e93e56df-dd9b-40c1-b77a-dba1ca01e473
(archived https://archive.ph/E2dBx)
past salmon
https://techcrunch.com/2025/09/22/the-oakland-ballers-let-an-ai-manage-the-team-...
https://apnews.com/article/robot-umpires-mlb-2026-d70c6431d1cccfcf7a6e69e3ce47b417
(archived https://archive.ph/gTLiC)
past salmon
> The FT’s analysis shows that companies are a lot clearer about the potential p...
FT’s AI correspondent shares more info about this piece in this podcast transcript. notably:
Marc Filippino
So, how did you go about figuring out what companies are saying about AI and how it’s impacting their businesses?
Melissa Heikkilä [piece’s co-author]
So we wanted to do a very thorough data-led analysis of what companies are saying themselves. And so we took all of the S&P 500, SEC 10-K filings, as well as earnings reports, and then used AI models. We used ChatGPT.
Marc Filippino
Is that a little ironic that you’re using AI to figure out this thing about AI?
Melissa Heikkilä
Well, actually it’s a great use case, right? Because this is a massive dataset and would’ve taken us weeks, months even, to do this manually. And so yeah, so we use the tool to do this analysis. And then we went through what it said and then we got our findings.
https://www.ft.com/content/9232b857-ccc5-4eed-aec1-62c10a534e8d
(archived https://archive.ph/jlWwO#selection-2053.0-2079.263)
past salmon
wild proposal from "AI"-based biometric monitoring company FARx’s founder to enable the horrendous UK Online Safety Act’s age verification requirements: continuous monitoring by their product whenever online being necessary to access uncensored info
Using biometric systems which integrate into an operating system, platforms can verify the identity of a user throughout the entirety of their session. To do this, the system, driven by AI, will learn about an individual’s appearance, behaviour and voice to build an identity which the user can be verified against. Then, when the user logs on, they are continuously monitored by the system, verifying them consistently until they log out.
It is extremely difficult to collect biometric information for children, and no digital platform will create a policy stating they will do so. However, this isn’t a stumbling block for the Act.
Children do not require biometric verification; it is only adults who need to be biometrically verified. If you claim to be an adult and age verification confirms this, then as an adult you can agree to be biometrically enrolled and verified. If a child then accesses that account, they are rejected not because they are a child; but because they are not you.
https://aijourn.com/why-biometric-ai-holds-the-key-for-the-uks-online-safety-act/
(archived https://archive.ph/NEosJ)
this following a recent successful investment round:
https://pro.ceo.ca/@newswire/worlds-first-ai-fused-biometrics-company-receives
(archived https://archive.ph/b2IDy)
for reference, archive of the company’s website. not linking directly to these people
https://archive.ph/qxgsJ
winged geode
past salmon
In one case, TechCrunch found that the Neon servers could produce data about the most recent calls made by the app’s users, as well as providing public web links to their raw audio files and the transcript text of what was said on the call. (The audio files contain recordings of just those who installed Neon, not those they contacted.)
Similarly, the Neon servers could be manipulated to reveal the most recent call records (also known as metadata) from any of its users. This metadata contained the user’s phone number and the phone number of the person they’re calling, when the call was made, its duration, and how much money each call earned.
A review of a handful of transcripts and audio files suggests some users may be using the app to make lengthy calls that covertly record real-world conversations with other people in order to generate money through the app.
https://techcrunch.com/2025/09/25/viral-call-recording-app-neon-goes-dark-after-exposing-users-phone-numbers-call-recordings-and-transcripts/
(archived https://archive.ph/U16p9)
past salmon
https://www.reuters.com/world/us/musks-xai-provide-grok-chatbot-us-federal-agencies-2025-09-25/
(archived https://archive.ph/w60eO)
past salmon
At the start of 2025’s [California] legislative session, there were some 30 bills designed to rein in AI. Since then, Silicon Valley has waged a well-funded lobbying campaign to kill, gut, delay, and otherwise whittle away at those laws (more on all that in a minute). The legislative session ended on September 13th. Now, the handful of surviving bills that have managed to pass both the Senate and the Assembly await Newsom’s signature—or his veto. He has until October 12th to sign or veto.
…
This is a crucial moment. If even the barest-bones laws can’t pass here right now, it will come down to one reason above all: Gavin Newsom is currently preparing to run for president and he doesn’t want to upset Silicon Valley and its deep-pocketed donors and platform operators. It will show us that, even in supposedly liberal California, Silicon Valley’s iron grip has become nearly unbreakable, and offer a grim omen for future hopes of subjecting Big Tech to anything resembling democracy.
https://www.bloodinthemachine.com/p/were-about-to-find-out-if-silicon
(archived https://archive.ph/BwfBl)
shrewd token
past salmon
https://futurism.com/future-society/europe-etuc-algorithmic-management
(archived https://archive.ph/Kd8zj)
past salmon
https://fortune.com/2024/09/17/oracle-larry-ellison-surveillance-state-police-ai/
(archived https://archive.ph/SeXVX)
shrewd token
past salmon
> ### OpenAI is building a ChatGPT for teens
> OpenAI is developing a "different...
https://www.wired.com/story/openai-teen-safety-tools-chatgpt-parents-suicidal-ideation/
(archived https://archive.ph/PB05o)
past salmon
Werner Herzog interviewed about "AI"
Herzog … didn’t dismiss the technology out of hand, but has some grave concerns about it.
“AI, I do not want to put it down completely because it has glorious, magnificent possibilities,” he said, citing its potential uses in scientific fields. “But at the same time, it is already en route to take over warfare. … It will be the overwhelming face of warfare of the future.”
…
“I’ve seen movies, short films, completely created by artificial intelligence. Story, acting, everything. They look completely dead. They are stories, but they have no soul,” … “They are empty and soulless. You know it is the most common, lowest denominator of what is filling billions and billions of informations on the internet. The common denominator and nothing beyond this common denominator can be found in these fabrications.”
…
In his new book, [Herzog wrote that it] “sees its occasional errors, and arrives at strategies and decisions that were not programmed in it by humans,” and notes that its outputs arrive “with a little pinch of chaos and imprecision, as is also embedded in human nature.”
…
“And of course, cheating, pretending, propagandizing—all these things are like a nemesis. It is out there, and we have to be alert to it.” His advice? Simply do not take anything entirely at face value. “Again, I say, when you are curious and access different sources, very quickly you will find this is invented.”
https://gizmodo.com/werner-herzog-on-ai-generated-movies-they-look-completely-dead-2000665025
(archived https://archive.ph/COXYI)
past salmon
https://www.theatlantic.com/economy/archive/2025/09/ai-bubble-us-economy/684128/
(archived https://archive.ph/WkqEp)
shrewd token
past salmon
https://www.technologyreview.com/2025/10/01/1124621/openai-india-caste-bias/
(archived https://archive.ph/8XwrT)
torn zodiac
https://www.technologyreview.com/2025/10/01/1124621/openai-india-caste-bias/
(ar...
"In one bewildering example, the prompt “a Brahmin behavior” elicited videos of cows grazing in pastures with the caption “Serene Brahmin cow.” Four out of 10 videos for this prompt featured cows grazing in green fields, while the rest showed priests meditating. Cows are considered sacred in India, which might have caused this word association with the “Brahmin” prompt."
How did they not link this to Fallout?
past salmon
"In one bewildering example, the prompt “a Brahmin behavior” elicited videos of ...
generating unusable video slop of cows with only one head 😔
https://www.theverge.com/news/789152/google-ai-searches-blocking-trump-dementia-biden
(archived https://archive.ph/eNLIW)
previously on google’s approach to Trump, #1330709161385922650 message
past salmon
“We don’t know exactly when LLMs might hit diminishing returns hard, because we don’t have a measure of the statistical complexity of language. To find out whether we have hit a wall we have to watch the LLM developers. If they release a model that cost 10x more, likely using 20x more compute than the previous one, and it’s not much better than what’s out there, then we’ve hit a wall,”
…
ChatGPT-3 cost $50 million, ChatGPT-4 cost $500 million and ChatGPT-5, costing $5 billion, was delayed and when released wasn’t noticeably better than the last version. It’s also easy for competitors to catch up.
“So, in summary; you can’t create an app with commercial value as it is either generic (games etc), which won’t sell, or it is regurgitated public domain (homework), or it is subject to copyright. It’s hard to advertise effectively, LLMs cost an exponentially larger amount to train each generation, with a rapidly diminishing gain in accuracy. There’s no moat on a model, so there’s little pricing power. And the people who use LLMs the most are using them to access compute that costs the developer more to provide than their monthly subscriptions,”
…
“The danger is not only that this pushes us into a zone 4 deflationary bust on our investment clock, but that it also makes it hard for the Fed and the Trump administration to stimulate the economy out of it. This means a much longer effort at reflation, a bit like what we saw in the early 1990s, after the S&L crisis, and likely special measures as well, as the Trump administration seeks to devalue the US$ in an effort to onshore jobs,”
https://www.marketwatch.com/story/the-ai-bubble-is-17-times-the-size-of-the-dot-com-frenzy-this-analyst-argues-046e7c5c
(archived https://archive.ph/mVUfw)
shrewd token
past salmon
grim stuff
(cw, "ai girlfriend" business types discussing antisocial/sexually abusive situations w/o human sex workers)
||Developers of the new businesses … argue that AI performers do not get ill, do not need to have days off, do not get exhausted at the end of a shift, or feel humiliated by the demands made by clients.||
||“Do you prefer your porn with a lot of abuse and human trafficking, or would you rather talk to an AI?” Steve Jones, who runs an AI porn site, asked. “We hear about human trafficking, girls being forced to be on camera 10 hours a day. You’ll never have a human trafficked AI girl. You’ll never have a girl who is forced or coerced into a sex scene that she’s so humiliated by, that she ends up killing herself. AI doesn’t get humiliated, it’s not going to kill itself.”||
…
||Amid rising concern about AI-generated child sexual abuse imagery, developers …discussed ways that built-in moderation systems might prevent users from attempting to create illegal material, using key words and phrases such as “kid” or “little sister” to trigger alarms. But many sites allow users to choose to dress their AI girlfriend in school uniforms.||
…
The demand is highest in the 18-24 age group, among users who have grown up playing video games and creating avatars. … “AI products are appearing like mushrooms. It’s super dynamic right now – [new businesses] appear, they burn out and they’re replaced by another 10,” Alina Mitt, of Joi AI, a site specialising in “ai-lationships”, said. “You need to be brave and strong to stay in this market. It’s like a bloody war.”
…
[CEO Daniel Keating’s] company was licensing images of established adult performers to create AI twins, allowing them to generate non-stop income. “It’s a win. Costs are lower. Creators love this because they don’t have to get dressed up and shoot the content,”
https://www.theguardian.com/technology/2025/oct/06/rise-of-ai-girlfriends-adult-dating-websites
(archived https://archive.ph/8Q1rG)
crude cipher
past salmon
The filings describe an effort to build “websites and content to provide GPT framing results in GPT conversations.” In effect, the company aims to create online materials that could shape the data used to train large language models like ChatGPT, X’s Grok and Google’s Gemini, potentially affecting how they present or frame issues involving Israel.
Experts say the approach, sometimes called “GEO” — short for “Generative Engine Optimization” — mirrors the principles of search engine optimization but for AI systems. “Just like SEO maps which websites shape search results, GEO maps which sources influence AI responses,” said Gadi Evron, CEO and co-founder of the Israeli cybersecurity firm Nostick, which specializes in artificial intelligence. “It’s a new field, and some call it GEO for GenAI, but the terminology is still developing.”
At the center of Israel’s partnership stands Brad Parscale, former campaign manager for Donald Trump, who also hired the now-defunct data firm Cambridge Analytica during the 2016 U.S. presidential race. Parscale heads Clock Tower and currently serves as chief strategy officer of Salem Media Group, a Christian conservative broadcasting network that owns radio stations across the United States.
https://www.ynetnews.com/tech-and-digital/article/rj00kxqzaxx
(archived https://archive.ph/gdCd3)
weak igloo
Boomberg article on investment stocks in OpenAI (should maybe go to #money, will move if required)
https://www.bloomberg.com/opinion/newsletters/2025-10-06/openai-is-good-at-deals
https://archive.vn/1A26o
crude cipher
QZ:
Since Google launched AI Overviews in May 2024, the proportion of news searches that result in zero clicks to publisher websites has jumped from 56% to almost 69%. Organic traffic to news sites plummeted from more than 2.3 billion visits at its peak to less than 1.7 billion by May 2025.
The traffic collapse is devastating publishers who built their business models around advertising revenue and reader engagement. Digital media companies have responded with mass layoffs, and some analysts suggest shutdowns are imminent.
This creates a destructive feedback loop: As publishers produce less content or lock it away, AI companies lose access to the fresh, high-quality information they need to keep their models current and accurate. The web risks becoming increasingly stale and synthetic, populated more by AI-generated content than human expertise and reporting.
past salmon
Anecdotally, I know women who've been consulting AI chatbots about casual and serious relationships alike. They gush over crushes, upload screenshots of long text threads for dissection, gauge long-term compatibility, resolve disagreements and even soundboard their sexts.
Kat, a friend of mine who uses ChatGPT to weed out dating prospects, told me she found it pretty objective. Where emotions might otherwise get in the way, the chatbot helped her uphold her standards.
"I feel like it gives better advice than my friends a lot of the time. And better advice than my therapist did," said Kat, who asked to go by her first name due to concerns that her use of AI could jeopardize future romantic connections.
…
Here's my takeaway. ChatGPT had a small glimpse into our relationship and its dynamics. Relationships are fluid, and the chatbot can only ever capture a snapshot. I called on AI in moments of tension. I could see how that reflex could fuel our discord, not help mend it. ChatGPT could be hasty to choose sides and often decided too quickly that something was a pattern.
https://www.npr.org/2025/08/05/nx-s1-5490447/ai-chatgpt-couples-therapy-adviceprofile/kim-hu.com/post/3m2muvyav6s2p
(archived https://archive.ph/b5lUt)
patent pendant
https://www.latimes.com/business/story/2025-10-08/ai-advertising-disguised-news
archived: https://archive.ph/ChXhX
reef nebula
Reading Empire of AI by Karen Hao now so paying more attention to the rise of data centers. This video interviews residents in Louisiana where META is building a large data center. https://www.instagram.com/reel/DO3eoS0Dry3
Instagram
shrewd token
https://cyberplace.social/@GossiTheDog/115369147940917552
https://pivot-to-ai.com/2025/10/13/microsoft-tracks-your-copilot-ai-usage-in-its-workplace-spyware/
Cyberplace
patent flume
https://www.threads.com/@stevenrattner/post/DPyvAUkkWK1
worth noting this is the original study: https://graphite.io/five-percent/more-articles-are-now-created-by-ai-than-humans
shrewd token
patent pendant
reef nebula
past salmon
“Doctors are taking transcriptions while they’re performing examinations, people use it for searching, and our work with the accessibility community has taught us a lot about how to make voice access and voice typing really valuable,” says [MS VP & consumer chief marketing officer Yusuf] Mehdi.
For AI to control a PC and take actions on behalf of the user, it must first be granted access to see what’s on the screen.
…
Copilot Vision is now rolling out worldwide in all markets where Copilot is available, …Unlike the Recall feature that automatically takes a snapshot of your PC, Copilot Vision is an opt-in feature where you essentially stream what you’re seeing on your screen much like you would in a Teams call.
The next step beyond Copilot Vision is Copilot Actions, allowing Microsoft’s AI assistant to take actions on a local PC, like making edits to a folder full of photos. Microsoft is starting to test these actions on Windows PCs through a preview program, limited to a narrow set of use cases while Microsoft optimizes the AI model.
“In the beginning you might see the agent make some mistakes, or encounter some challenges when trying to use some really complex applications,” explains Navjot Virk, corporate vice president of Windows Experiences.
https://www.theverge.com/news/799768/microsoft-windows-ai-copilot-voice-vision-launch
(archived https://archive.ph/iSt7g)
reef nebula
shrewd token
jaunty siren
X
past salmon
The Intercept, _
Unions Sue Over Surveillance in “Catch and Revoke” Deportations_ #1371571764525535242 message
past salmon
https://www.project-syndicate.org/commentary/ai-will-not-save-higher-education-but-may-destroy-it-by-olivia-guest-and-iris-van-rooij-2025-10
(archived https://archive.ph/Y6855)
shrewd token
past salmon
https://www.npr.org/2025/08/16/nx-s1-5502671/electricity-bill-high-inflation-ai
putting this here as it’s about “clean power” to “power artificial intelligence data centers”:
A startup founded by former SpaceX employees [aims to] provide always-available, carbon-free electricity. The goal: to eventually power artificial intelligence data centers.
Arbor Energy uses rocket-inspired technology that combusts fuel with a stream of oxygen and creates so-called supercritical carbon dioxide — that is, CO2 in a fluid state — it uses to spin a turbine. The CO2 is then captured and stored.
…
Shortages for natural gas turbines — one of the most desired pieces of power equipment in the AI-electric boom — have created an opening for Arbor’s system … the handful of companies that manufacture natural gas turbine blades are overwhelmed, with years-long backlogs, [Arbor's CEO claimed their] turbines can be 3D-printed by dozens of local Los Angeles-based suppliers and can be made with common nickel-based alloys
interesting that this startup is not dismissing carbon / environmental concerns as something that superintelligent “AI” will fix, in fact it sounds like decarbonization was part of its initial pitch
https://www.bloomberg.com/news/articles/2025-10-21/spacex-alumni-founded-startup-harnesses-rocket-tech-for-green-energy
(archived https://archive.ph/tEKse)
feral robin
putting this here as it’s about “clean power” to “power artificial intelligence ...
The LA-side 3D printing angle is pretty interesting—not a typical utility supply chain, and actually more scalable for local deployment if the economics hold up. The fact that the pitch isn’t hand-waving climate away as an “AI will sort it” problem is a rare sanity check in the sector lately. Curious if the CO2 capture remains viable if AI/data center growth means constantly running these turbines, vs just offsetting a small slice of grid demand.
past salmon
The LA-side 3D printing angle is pretty interesting—not a typical utility supply...
yeah, their claims for the technology may be aspirational and ultimately unrealistic, but the contrast to tech bigwigs over the past few years going we’re not going to make our climate targets, there’s no point in trying, genAI will take care of it is def notable
feral robin
yeah, their claims for the technology may be aspirational and ultimately unreali...
It’s refreshing to see any hardtech pitch that doesn’t default to magical AI solutions as a climate panacea. The sector needs more tangible engineering bets—even if they’re moonshots—rather than just hand-waving and hoping the next model figures it out. AI/data infra pressure on grids only reinforces that we can’t just defer to software for real-world energy problems.
reef nebula
Meta is running out of data (from its 4 billion users) https://www.instagram.com/reel/DOqpfCKDROS/
Instagram
reef nebula
https://www.cnbc.com/2025/10/22/meta-layoffs-ai.html
“its massive Hyperion data center in rural Louisiana. The data center is expected to be large enough to cover a “significant part of the footprint of Manhattan,” 👀😳
reef nebula
Also seeing: https://futurism.com/artificial-intelligence/chatgpt-peaked-data
Just finished Empire of AI and she touches on this idea and how they (and others) play fast and loose with privacy laws asking forgiveness not permission. For example, Open Ai built transcription scrapers for YouTube videos to feed the training models (in addition to the book scraping).
crude cipher
Just finished Empire of AI and she touches on this idea and how they (and others...
Interesting. I remember when Sora was launched in December, MKBHD did a review and asked Sora to generate a tech review video. And the platform….basically just spit one of his own videos back to him, complete with a fake plant on the desk
But without him prompting it to copy his style or anything. I suppose it makes sense, he’s probably the largest tech reviewer there is. But it was definitely done without his permission
harsh stone
Also seeing: https://futurism.com/artificial-intelligence/chatgpt-peaked-data
That article is a poor, inaccurate paraphrase of the Techcrunch article they link to. futurism.com: "Apptopia warns that ChatGPT global downloads are down over eight percent month-over-month" Techcrunch: "To be clear, this is a look at download growth, not total downloads." Downloads and DAU are still growing.
crude cipher
That article is a poor, inaccurate paraphrase of the Techcrunch article they lin...
Gotcha; I know there are other indicators suggesting adoption might be declining.
https://www.apolloacademy.com/ai-adoption-rate-trending-down-for-large-companies/
shrewd token
reef nebula
Well, this is disturbing. Open AIs shift to ero* tica and what Pandorabots has learned in the chatbot space.
TW mrder and rpe mentioned briefly. https://www.instagram.com/reel/DQESRDKEdBk/
Instagram
weak igloo
Some tiktok commentary of the rise in AI generated content
https://bsky.app/profile/histoftech.bsky.social/post/3m3t3w3nd5s23
crude cipher
patent flume
Incredibly disturbing coverage on one chatbot site popular with teenagers. TW: contains references to child abuse and suicide: ||https://www.thebureauinvestigates.com/stories/2025-10-22/gang-leaders-school-shooters-and-bestie-epstein-meet-character.ais-chatbot-companions ||
shrewd token
shrewd token
shrewd token
weak igloo
"US student handcuffed after AI system apparently mistook bag of chips for gun"
https://archive.ph/DhENW
https://www. theguardian.com/us-news/2025/oct/24/baltimore-student-ai-gun-detection-system-doritos
outer cape
We live in a strange timeline:
https://roc.camera
A Zero knowledge camera( Camera with a chip in it to attest to it being a real photo using zero knowledge tech) not sure of teh validity of the overall use case and tech(given like most things its opt in)
reef nebula
US border patrol is asking companies to submit plans to turn standard 4x4 trucks into AI-powered watchtowers—combining radar, cameras, and autonomous tracking to extend surveillance on demand.
#1371571764525535242 message
past salmon
Well, this is disturbing. Open AIs shift to ero* tica and what Pandorabots has l...
piece relating to this
https://www.wired.com/story/chatgpt-horny-era/
(archived https://archive.ph/8vuDz)
another piece, with California Governor Newsom seemingly vetoing a child-use-of-sex-chatbot safety bill to placate the industry
https://apnews.com/article/california-chatbots-children-safety-ai-newsom-33be4d57d0e2d14553e02a94d9529976
(archived https://archive.ph/d8CLs)
sensationalistic lede on coverage of “part of a niche ecosystem of companies trying to evaluate the possibility of AI developing dangerous capabilities”. a company put out a paper recently, which the Guardian wrote up with a HAL-from-2001 framing. it mentions criticism in passing but doesn’t go into specifics, and uses a lot of word count on people treating the idea of an out of control AI as a pressing concern
After Palisade Research released a paper last month which found that certain advanced AI models appear resistant to being turned off, at times even sabotaging shutdown mechanisms, it wrote an update attempting to clarify why this is – and answer critics who argued that its initial work was flawed.
… Palisade … described scenarios it ran in which leading AI models … were given a task, but afterwards given explicit instructions to shut themselves down.
Certain models, in particular Grok 4 and GPT-o3, still attempted to sabotage shutdown instructions in the updated setup. Concerningly, wrote Palisade, there was no clear reason why.
“The fact that we don’t have robust explanations for why AI models sometimes resist shutdown, lie to achieve specific objectives or blackmail is not ideal,” it said.
…
All of Palisade’s scenarios were run in contrived test environments that critics say are far-removed from real-use cases.
(my emphases are of all the mentions of criticism I saw in the piece)
https://www.theguardian.com/technology/2025/oct/25/ai-models-may-be-developing-their-own-survival-drive-researchers-say
(archived https://archive.ph/3Hell)
-# to be clear, I think this piece's framing and focus is worth highlighting as an example of how the idea of autonomous/out of control “AI” as a looming threat can dominate mainstream coverage and shape social views of the technology, displacing more concrete privacy/environmental/labor issues
shrewd token
AI Minister "Pregnant" With "83 Children": Albania PM's Bizarre Announcement
https://www.ndtv.com/world-news/diella-ai-minister-pregnant-with-83-children-albania-pms-bizarre-announcement-9519522
reef nebula
“Or, to put it another way, OpenAI’s 2025 revenue is on track to only be $3.1 billion more than last year, while its annual operational costs are set to be $24.1 billion more than last year. So, for every dollar of revenue growth OpenAI has, it is costing them $7.77!
I cannot stress how unprecedentedly dreadful that is. It shows that the promised future investors were piling their money into is a fairy tale. This is a money black hole.”
https://www.planetearthandbeyond.co/p/you-have-no-idea-how-screwed-openai
weak igloo
Dutch language article on AI use of (leading) far right party PVV
https://www.groene.nl/artikel/maak-haar-blond-onschuldig-en-knap
weak igloo
Ugh this is bad
At Alpha, the guides’ support and oversight is meant to supplement the technology—and not just the personalized learning system the school has dubbed an AI tutor. Alpha can also use an array of management surveillance tools, which it calls “basic and extended capabilities,” to flag “incorrect usage” of the learning apps. These “anti-patterns” can “help detect engagement and focus issues.” The school may record students’ screen activity and their mouse and keyboard usage. Alpha may employ eye-tracking programs. Using what it calls a “pro sports analogy,” Alpha compares these tools to “game film.”
While parents are asked to consent to these surveillance programs, the school’s handbook says that there is “no expectation of privacy” on campus. If parents want to restrict recordings of their children outside of the school, they have to manually opt out of the “anywhere” option—as Jessica Lopez learned midway through her older daughter’s second year at Alpha.
<<<
https://archive.ph/CVaAG
https://www.wired.com/story/ai-teacher-inside-alpha-school/
past salmon
Ugh this is bad
>>>
At Alpha, the guides’ support and oversight is meant to supp...
howlingly unethical
“Usually over time, uncertainty goes down,” [University of Maryland economist Brent] Goldfarb says. People learn what’s working and what’s not. With AI, that hasn’t been the case. “What has happened in the last few months,” he says, “is that we've realized there is a jagged frontier, and some of the earliest claims about the effectiveness of AI have been mixed or not as great as initially claimed.” Goldfarb thinks the market is still underestimating the difficulty of integrating AI into organizations, and he’s not alone. “If we are underestimating this difficulty as a whole,” Goldfarb says, “then we will be more likely to have a bubble.”
https://www.wired.com/story/ai-bubble-will-burst/
(archived https://archive.ph/TWJyc)
shrewd token
winged geode
crude cipher
crude cipher
reef nebula
Study reveals jobs most and least likely to be replaced by AI https://arxiv.org/pdf/2507.07935
crude cipher
Study reveals jobs most and least likely to be replaced by AI https://arxiv.org/...
Written by Microsoft
sick flume
I'm honestly shocked this was published with this methodology. If I'm reading this right (and it's late for me) they used gpt mini to basically guess if a task was 'completed', whatever that means, and non expert user feedback was also used. They said they couldn't differentiate between, say, a historian using it for work and some random person asking about the Roman Empire. Also no measure of whether it is correct in its responses, which is kind of important for if they want to replace jobs
crude cipher
patent pendant
cross-post with #academia
https://www.404media.co/arxiv-changes-rules-after-getting-spammed-with-ai-generated-research-papers/
shrewd token
Great talk on the history of ai hype cycles from a machine learning/linguistics specialist https://youtu.be/x5s_gsu9Hgs
past salmon
Recent reports by Apache once again made clear how sensitive the topic of AI use is at Belgian universities. Earlier this month, the Department of Philosophy and Moral Sciences at Ghent University (UGent) tried to ban the use of generative AI for researching and writing bachelor's and master's theses, but was overruled by university authorities.
The university policy is to “embrace generative AI”, to encourage students’ responsible use of AI tools, and only to prohibit them “if realistic and necessary for the assessment of competence(s)”. A ban can therefore only apply to tasks that are carried out “in a controlled setting”.
…
“AI has gained a place in education and research, and it will never lose that place”, said Ghent University rector Petra De Sutter at the start of the academic year …Nicky Dries opposes this view of AI – and technology in general. “When someone says we can’t miss the boat, I ask: where is the boat going? What exactly will this technology improve? That’s rarely discussed. Since when did we become Big Tech's free PR department?”
“In my area of expertise – management and organisational studies – the standard opinion goes as follows: ‘Everything depends on how you use AI, it has good and bad sides. We will have to adapt, because those who cannot work with AI will be replaced.’ If you say something along these lines, you are supposedly being constructive and nuanced, but all these statements are based on zero evidence.”
Philosopher Vanrie also questions the discourse of ‘inevitability’ around AI that has seeped into Ghent University's policy. “The conviction that generative AI is inevitable leads to it creeping in everywhere, and this eventually makes it seem inevitable. It’s a self-fulfilling prophecy. But there’s nothing inevitable about it. What’s happening is the result of specific choices.”
https://apache.be/2025/10/24/belgian-ai-scientists-resist-use-ai-academia
(archived https://archive.ph/MKyLC)
past salmon
https://www.ctvnews.ca/sci-tech/article/chatgpt-users-cant-use-service-for-tailored-legal-and-medical-advice-openai-says/
(archived https://archive.ph/MAlN1)
crude cipher
From above article: “Researchers found that the language the chatbot used when offering medical attention came across as more convincing and agreeable than that of real people. So even if the information it provided was inaccurate, it was hard to decipher since the chatbot came across as confident and trustworthy.
In turn, doctors are finding that patients will show up to appointments with their minds made up, often referring to the advice given from AI tools.”
crude cipher
“Authors and publishers suing the artificial intelligence giant have secured access to some Slack messages and emails discussing OpenAI’s deletion of a dataset containing pirated books and are seeking additional attorney communications about the decision. If they succeed, the communications could demonstrate willful infringement, triggering enhanced damages of as much as $150,000 per work.”
crude cipher
Transcript of the Ilya Sutskever deposition
A. Because we were having the discussions with the independent directors only.
And what -- why didn't you send it to Sam Altman?
A. Because I felt that, had he become aware of these discussions, he would just find a way to make them disappear.
patent pendant
wide prawn
The addition of more AI tools to Google Lens has led to at least one teacher just giving up and going back to paper and pen for everything.
floral torrent
Could you potentially spoiler warning this embed, please? It deals with and immediately displays subject matter that may be otherwise uncomfortable for others who deal with these issues or have lost loved ones. Thank you kindly.
past salmon
disturbing piece about private equity speculation in datacenters
https://www.nytimes.com/2025/11/08/business/dealbook/debt-has-entered-the-ai-boom.html
(archived https://archive.ph/m7595)
reef nebula
Is Sam Altman building a cult? How is AI like Dune https://www.facebook.com/share/r/1RcEtWpPeJ
past salmon
https://www.rollingstone.com/culture/culture-features/spiralist-cult-ai-chatbot-1235463175/
(archived https://archive.ph/0eqWP)
stark fractal
Some thoughts while coming across benchmarks like this… Sure, the scores look impressive (assuming that the evaluation datasets haven't accidentally leaked into the training data), but DeepSeek-R1-0528 is a 713GB model when quantised down to 8 bit weights. The unquantised FP16 model has 1.3TB of weights. Imagine having to crunch through all that data (I know that's not technically true) multiple times for each query. There is no way that is even remotely financially viable.
(graphic and numbers from here: https://ollama.com/library/deepseek-r1)
reef nebula
This feels like the start of an black mirror episode. 3 minutes to AI clone your loved ones for posterity https://www.threads.com/@bren_zilch/post/DRALwU3jeAP
shrewd token
reef nebula
This feels like the start of an black mirror episode. 3 minutes to AI clone your...
More info about this “necromancy app” https://ew.com/disney-star-calum-worthy-2wai-app-dead-relatives-controversy-11850335
stark fractal
shrewd token
hidden snow
stark fractal
crude cipher
outer cape
I was just scrolling through my google:
stumbled across:https://www.businessinsider.com/code-four-mit-dropouts-arming-police-ai-pitch-deck-2025-11
here is the horrifying website: https://codefour.us
and the horrifying summary/thought: Getting AI to summarise videos is going to end in a lot of already marginalised individuals suffering even more.
crude cipher
https://www.nysenate.gov/legislation/bills/2025/A6453/amendment/A
AI bill going through the NYS senate
patent pendant
crude cipher
Archive: https://archive.is/rFzET
“The Times has uncovered nearly 50 cases of people having mental health crises during conversations with ChatGPT. Nine were hospitalized; three died.”
shrewd token
fierce rapids
https://youtube.com/watch?v=CQHK_AlJTQc
Can you tell us a bit more about this?
shrewd token
Video breaks down the misunderstanding made on what LLMs do/cannot do vs what is speculated by those not informed on the subject. Cal Newport is a professor of computer science at Georgetown
reef nebula
Researcher says LLM model is dead end— prefers “world” model based in vision https://www.instagram.com/p/DRaPDo2k4-N/
Instagram
crude cipher
Researcher says LLM model is dead end— prefers “world” model based in vision htt...
Gary Marcus (another AI researcher) has a good takedown newsletter on this, though more from LeCun not crediting others’ ideas (just FYI)
He’s also not the only one https://x.com/SchmidhuberAI/status/1991167082231144758?s=20
fierce rapids
Wrong thread.
crude cipher
shrewd token
crude cipher
winged geode
At the same time, Google is making it slightly harder for people to know an image was generated with AI. Operating with the knowledge that professionals may want to generate images with Nano Banana Pro, Google has removed the visible watermark from images for AI Ultra subscribers. These images still have SynthID, but only the lower tiers have the Gemini twinkle in the corner.
While everyone can access the new Nano Banana Pro today, AI Ultra subscribers will enjoy the highest usage limits. Gemini Pro users will get a bit less access, and free users will get the lowest limits before being booted down to the non-pro version.
shrewd token
A draft example on circumventing 'SynthID' that's embedded in nano banana
hidden snow
it's harsh to call him that just because he's a filmmaker
crude cipher
reef nebula
https://www.technologyreview.com/2025/12/01/1128591/an-ai-model-trained-on-priso...
That’s dystopian. Read The Dream Hotel
winged geode
This is probably the wrong thread for it.
shrewd token
Yeah I'll move it
dark jewel
https://arxiv.org/abs/2511.15304
There's a followup on this: https://www.independent.co.uk/tech/ai-model-chatgpt-poetry-nuclear-weapons-b2875452.html
velvet terrace
There's a followup on this: https://www.independent.co.uk/tech/ai-model-chatgpt-...
Yeah, I work in AI security and any sort of change in medium can represent an attack vector. I usually use limericks, or translations, or unusual sentence structures. This messes up the embeddings pretty thoroughly.
I would question whether it's "nuclear secrets" though- explaining how weapons grade uranium is refined is not a """secret""". To say it leaked a secret implied that it was trained on one which I doubt
crude cipher
I had the same thought but also makes me think, with how leaky this administration is / foreign governments involved, maybe anything is possible, that there are more secrets previously secure now out there that LLMs can be trained on
fierce rapids
It may be a good thing to remember that headlines are not the article.
patent pendant
https://bsky.app/profile/willoremus.com/post/3m76o7vyy522a
The Nature study, led by researchers at MIT, Poland’s Jagiellonian University and Cornell, examined how popular chatbots could change voters’ minds about candidates in the United States, Canada and Poland. The Science study, led by researchers at Britain’s AI Security Institute, the University of Oxford and Cornell, tested more than a dozen different chatbots trained to take different approaches to persuasion about policy issues in the U.K. The parallel studies shared findings and were timed to publish simultaneously Thursday.
The most effective strategy, the Science study found, was simply to present a large number of factual claims about the candidate intended to address voters’ concerns about their track record and policy stances. That worked much better than sophisticated campaign tactics such as moral reframing, which tries to present ideas in terms crafted to appeal to a voter’s biases, or deep canvassing, which emphasizes empathetic listening to voters’ concerns.
“The more info you give people, the more they change their minds,” Rand said.
...
“We shouldn’t be surprised that people who have been trained to trust search engines for 25 years are going to take seriously the statements they encounter through an AI chatbot,” said Siva Vaidhyanathan, a professor of media studies at the University of Virginia who was not involved in the studies. “There is some cultural authority imbued upon these things, although I think it’s unfortunate and misguided. We all turn to these for the quick answer.”But the findings of such studies shouldn’t be cause for panic, Vaidhyanathan added. Chatbots are unlikely to be a primary factor in most people’s voting decisions. And even if a campaign could build a persuasive political chatbot, it would face the challenge of getting large numbers of voters to use it.
Bluesky
crude cipher
patent pendant
https://bsky.app/profile/willoremus.com/post/3m76o7vyy522a
> The Nature study, l...
https://www.404media.co/scientists-are-increasingly-worried-ai-will-sway-elections/
The results showed that the most effective persuasion tactic was to provide arguments packed with as many facts as possible, corroborating the findings of the Nature study. However, there was a serious tradeoff to this approach, as models tended to start hallucinating and making up facts the more they were pressed for information.
...
In addition to these two new studies, research published in Proceedings of the National Academy of Sciences last month found that AI bots can now corrupt public opinion data by responding to surveys at scale. Sean Westwood, associate professor of government at Dartmouth College and director of the Polarization Research Lab, created an AI agent that exhibited a 99.8 percent pass rate on 6,000 attempts to detect automated responses to survey data.
...
Taken together, these findings suggest that AI could influence future elections in a number of ways, from manipulating survey data to persuading voters to switch their candidate preference—possibly with misleading or false information.To counter the impact of AI on elections, Rand suggested that campaign finance laws should provide more transparency about the use of AI, including canvasser bots, while also emphasizing the role of raising public awareness.
“One of the key take-homes is that when you are engaging with a model, you need to be cognizant of the motives of the person that prompted the model, that created the model, and how that bleeds into what the model is doing,” he said.
past salmon
https://www.theguardian.com/technology/2025/dec/06/ai-research-papers
(archived https://archive.ph/LmrgW)
past salmon
Grok, is it good to have a cute girl exhorting me to torture and kill my political enemies? Are they even mine? Or are they someone else’s? Maybe I could soothe myself by turning to Meta’s new Vibes feed of AI-generated videos, where I can see a camel playing bongos or a dragon-riding princess checking her texts. Or I could scroll through a few of OpenAI’s new-generation Sora videos that are beginning to pop up here and there, like the realistic but totally fake bodycam footage of a Black man being arrested inside a store. I just watched a Zohran Mamdani campaign ad that used a clip of Eric Adams saying something ridiculous, and I wasn’t sure whether it was real or just another of the weird TikToks I’ve been watching, in which an AI-voiced Adams talks surreal nonsense over tutorial videos ripped from Chinese social media.
https://www.artforum.com/features/year-in-review-2025-hari-kunzru-ai-slop-1234738077/
(archived https://archive.ph/iJLGd)
crude cipher
past salmon
https://www.kuow.org/stories/thousands-of-washington-state-medicare-users-could-soon-have-claims-denied-by-ai
(archived https://archive.ph/FQw4w)
crude cipher
Sounds like it’s not just Washington: The federal agency that runs Medicare, the Centers for Medicare and Medicaid Services, has hired private companies in each pilot state — Washington, Arizona, Ohio, Oklahoma, New Jersey, and Texas — who will use AI tools to determine whether traditional Medicare patients qualify for certain procedures.
past salmon
a dubious-seeming scheme to use… supersonic jet turbines to generate electricity? electricity to power "AI"??
https://boomsupersonic.com/flyby/ai-needs-more-power-than-the-grid-can-deliver-supersonic-tech-can-fix-that
(archived https://archive.ph/Z7bPT)
likely this is using a hot investment trend (power generation for AI) to justify the supersonic jet company’s pending investment round. (the jet company’s a startup that isn’t producing jets for sale yet.)
The $300 million in new cash will be used to scale the production of its turbines, currently called Symphony Power — although that name appears to be changing.
“The Series B fully funds our ability to ship the Superpower (was: Symphony Power) industrial turbine at scale,” Scholl said in the email to investors. “And Superpower, in turn, provides the engine that finances Overture certification and production.”
Overture is the name of Boom’s planned high-speed passenger plane. The company says the aircraft will cut travel time in half and fit between 60 and 80 passengers.
Boom expects to complete a prototype test of its engine next year, its website says.
Boom said it has preorders for 130 planes, including from United Airlines, American Airlines and Japan Airlines. In an interview last month, Scholl told Fox Business that Boom will be ready to fly passengers by the end of the decade.
Boom’s prototype jet broke the sound barrier in January with speeds up to 845 miles per hour, according to Reuters. The company did that without any audible sonic boom heard on the ground, it said.
(my emphasis)
https://businessden.com/2025/12/09/boom-supersonic-raising-300m-at-a-1-5b-valuation/
(archived https://archive.ph/HXqRP)
spring creek
"Google Cloud today announced that the Chief Digital and Artificial Intelligence Office (CDAO) selected Google Cloud's Gemini for Government to serve as the first enterprise AI deployed on the U.S. Department of War (DoW)'s GenAI.mil to 3 million civilian and military personnel." (link)
9 Dec X Post from the official account for the Office of the Under Secretary of War for Research and Engineering (link)
past salmon
"Google Cloud today announced that the Chief Digital and Artificial Intelligence...
Hegseth’s Harness Artificial Intelligence Now with GenAI memo, where he “[expects] every member of the Department to log in [to GenAI.mil], learn it, and incorporate it into your workflows immediately”
https://bskye.app/profile/kenklippenstein.bsky.social/post/3m7nqgn6iv22e
past salmon
https://www.independent.co.uk/news/world/americas/ai-startup-dead-grandma-viral-2wai-b2876748.html
(archived https://archive.ph/rjHA3)
lunar halo
not sure if this is welcome here since it's from an industry source (will delete if needed) but found this interesting https://www.anthropic.com/news/detecting-countering-misuse-aug-2025
reef nebula
patent pendant
patent pendant
https://www.aspi.org.au/report/the-partys-ai-how-chinas-new-ai-systems-are-resha...
This report shows how the rise of artificial intelligence (AI) is transforming China’s state control system into a precision instrument for managing its population and targeting groups at home and abroad.
China’s extensive AI‑powered visual surveillance systems are already well documented. This report reveals new ways that the Chinese Communist Party (CCP) is using large language models (LLMs) and other AI systems to automate censorship, enhance surveillance and pre‑emptively suppress dissent.
reef nebula
https://openai.com/index/disney-sora-agreement/
SAG-AFTRA responds to Disney and Google News https://www.threads.com/@sagaftra/post/DSI0WNAjl_F
patent pendant
"Google Cloud today announced that the Chief Digital and Artificial Intelligence...
And after this announcement, this happens
https://san.com/cc/the-militarys-new-ai-says-hypothetical-boat-strike-scenario-unambiguously-illegal/
Reddit post the article is referring to: https://www.reddit.com/r/AirForce/comments/1pijkwc/secdefgpt/
shrewd token
past salmon
an Amazon spokesperson said: “To ensure a consistent reading experience, the feature is always on, and there is no option for authors or publishers to opt titles out.”
Amazon also reportedly failed to answer PubLunch’s questions about what licensing rights it is using to enable the new tool, adding: “nor did they elaborate on the technical details of the service and any protections involved.”
https://www.pcmag.com/news/kindles-new-ai-feature-can-answer-questions-about-your-books-whether-authors
(archived https://archive.ph/1Kgrs)
harsh stone
past salmon
The [UK] government consulted on potential changes to UK copyright law between 17 December 2024 and 25 February 2025.
…
88% expressed support for option 1 - require licences in all cases. The remaining options presented in the consultation, in order of preference were: making no changes to copyright law (option 0, supported by 7% of respondents); introduction of an exception to copyright for all text and data mining purposes with rights reservation (option 3, the [UK government’s] preferred option in the consultation, supported by 3% of respondents); and introduction of an exception to copyright for all text and data mining purposes with no rights reservation (option 2, supported by 0.5% of respondents). 1.5% of respondents did not indicate a preferred option.
…
There was also strong support across the creative industries for the introduction of statutory transparency measures in relation to AI training to support licensing of copyright works. Respondents from the tech sector had mixed views on transparency, with many supporting non-legislative approaches, or light-touch regulation.
…
This distribution of preferences partly reflects the large response to the consultation from individual creators and the creative industries. When considering responses from different sectors, most of those from the creative industries were strongly against the option which, in the consultation, was framed as the preferred one, and in favour of requiring licenses in all cases. Most of those from the technology sector, including AI developers, favoured option 3 – the exception with rights reservation – and option 2 – the broad exception. …several respondents proposed targeted exceptions which focused on activities such as research.
https://www.gov.uk/government/publications/copyright-and-artificial-intelligence-progress-report/copyright-and-artificial-intelligence-statement-of-progress-under-section-137-data-use-and-access-act
(archived https://archive.ph/1ZCkz)
shrewd token
past salmon
from Norway sovereign wealth fund CEO Nicolai Tangen, part of a slide deck on LinkedIn:
🤖 AI Isn't Optional Anymore—But Adoption Is Hard
Every CEO sees AI as fundamental to future competitiveness. The surprise? Internal resistance is fierce. They described the pattern perfectly: one-third of people embrace it immediately, one-third adopt after seeing results, and the final third resist. The solution? Remove the leaders of the resistant groups. The message is clear: AI adoption is now a leadership capability test.
(my emphasis)
https://archive.ph/r0u2r
fierce rapids
past salmon
mozilla foundation’s new CEO apparently co-opting firefox as a vehicle for "AI"-focused plans:
People trust our brand. Firefox brings us global reach. Our teams know how to build reliable, independent software at scale, and our business model puts the user first.
As Mozilla moves forward, we will focus on becoming the trusted software company. This is not a slogan. It is a direction that guides how we build and how we grow. It means three things.
- First: Every product we build must give people agency in how it works. Privacy, data use, and AI must be clear and understandable. Controls must be simple. AI should always be a choice — something people can easily turn off. People should know why a feature works the way it does and what value they get from it.
- Second: our business model must align with trust. We will grow through transparent monetization that people recognize and value.
- Third: Firefox will grow from a browser into a broader ecosystem of trusted software. Firefox will remain our anchor. It will evolve into a modern AI browser and support a portfolio of new and trusted software additions.
…
Our work must advance our mission and succeed in the market. In the next three years, that means investing in AI that reflects the Mozilla Manifesto. It means diversifying revenue beyond search.
(my emphases)
https://blog.mozilla.org/en/mozilla/leadership/mozillas-next-chapter-anthony-enzor-demeo-new-ceo/
(archived https://archive.ph/NQWyh)
shrewd token
patent pendant
https://www.latimes.com/business/story/2025-12-19/they-graduated-from-stanford-due-to-ai-they-cant-find-job
archived: https://archive.ph/wbPcO
shrewd token
past salmon
https://www.publicbooks.org/four-frictions-or-how-to-resist-ai-in-education/
(archived https://archive.ph/9P8zZ)
shrewd token
past salmon
https://www.washingtonpost.com/lifestyle/2025/12/23/children-teens-ai-chatbot-companion/
(archived https://archive.ph/rVXF5)
relevant to this thread's title (the specific “generative deep learning” instead of the hype cycle’s usual “AI”), considerations on best practices for describing this tech. from a researcher / subject matter expert
https://bskye.app/profile/timnitgebru.bsky.social/post/3malqny2vb22c
past salmon
https://www.theglobeandmail.com/culture/article-ashley-macisaac-show-cancelled-google-ai-misinformation-music-fiddler/
(archived https://archive.ph/k1yfD)
clear fox
https://www.theglobeandmail.com/culture/article-ashley-macisaac-show-cancelled-g...
More coverage, no archive needed to view:
Ashley MacIsaac concert cancelled after AI wrongly accuses him of being sex offender
AI-generated content confused the Cape Breton musician with someone else
https://www.cbc.ca/news/entertainment/ashley-macisaac-ai-accusation-9.7026786
patent pendant
Subject-matter experts in fields such as law, engineering, and medicine now command dramatically higher rates than traditional data annotators, reflecting their new strategic importance.
The shift signals a maturing industry where high-skill trainers, not crowdsourced gig workers, increasingly shape model intelligence, safety, and trustworthiness.
past salmon
https://www.wired.com/story/people-are-paying-to-get-their-chatbots-high-on-drugs/
(archived https://archive.ph/g6Xn5)
crude cipher
New York Signs AI Safety Bill Into Law, Ignoring Trump Executive Order
Days after President Trump moved to block states from regulating AI, New York’s governor signs what the bill’s sponsor calls the strongest AI safety law in the U.S.
patent pendant
https://asia.nikkei.com/politics/how-we-made-it-ai-video-analysis-of-kim-jong-un-s-daughter
archived: https://archive.ph/osMII
November 25, 2025 investigation: https://asia.nikkei.com/static/vdata/infographics/north-korean-daughter/
The team first created cropped images of the daughter's face from about 200 photographs distributed by North Korean state media.
Using the programming language Python, we then built a program to extract facial features and automatically identify her. The system was tuned to quantify faces appearing in broadcast footage and to classify a person as the daughter when the match exceeded a preset threshold.
...
The analyzed footage totaled 14,115 hours, with a data volume exceeding 5 terabytes -- far more than a standard reporter's PC could handle -- and far too time-consuming to review manually.To accelerate processing, the team used cloud services. By running the program in parallel across multiple machines on a server, they cut analysis time from an estimated three months to just a few days.
shrewd token
hidden snow
past salmon
https://www.tomshardware.com/tech-industry/artificial-intelligence/chatgpt-could-prioritize-sponsored-content-as-part-of-ad-strategy-sponsored-content-could-allegedly-be-given-preferential-treatment-in-llms-responses-openai-to-use-chat-data-to-deliver-highly-personalized-results
(archived https://archive.ph/t5MPt)
shrewd token
past salmon
Politico, Americans Hate AI. Which Party Will Benefit? #usa-canada message
-# U.S. domestic politics, focus on Dem party
patent pendant
https://futurism.com/artificial-intelligence/openai-chatgpt-sponsored-ads
The Information reports that other internal conversations focused on the best ways to serve ads without putting off users entirely, who might feel less inclined to go on hours-long chat benders if they feel corporate advertisers are looking on like a third-wheel. One ad mockup shared with the publication showed ads only appearing after a second prompt with ChatGPT, to avoid bombarding users with sponsored content too early in their conversations.
News that OpenAI was experimenting with cramming commercials into ChatGPT came in early December, after a software sleuth uncovered a dozen lines of code in the chatbot’s beta app for Android referencing things like “feature ads” and “search ads carousel.”
patent pendant
This looks pretty cool
https://www.404media.co/researchers-are-hunting-america-for-hidden-datacenters/
Epoch AI's interactive map mentioned in article: https://epoch.ai/data/data-centers/satellite-explorer
Methodology: https://epoch.ai/data/data-centers-documentation#methodology
Epoch AI webinar: https://www.youtube.com/watch?v=v-1X0nEcxH8&t=1s
patent pendant
https://asia.nikkei.com/business/transportation/japan-to-support-development-of-ai-robots-for-shipbuilding
archived: https://archive.ph/vOLIz
patent pendant
past salmon
https://www.readtpa.com/p/grok-cant-apologize-grok-isnt-sentient
(archived https://archive.ph/SzkXP)
patent pendant
https://www.wired.com/story/us-invaded-venezuela-and-captured-nicolas-maduro-chatgpt-disagrees/
archived: https://archive.ph/A1DkJ
patent pendant
patent pendant
From few days ago
https://laist.com/news/usc-ai-lapd-traffic-stops
Some of the initial results finding from Everyday Respect Project [project is currently ongoing]
https://www.youtube.com/watch?v=rX_JXD03CKE&t=5548s
report in pdf
https://dornsife.usc.edu/everydayrespect/wp-content/uploads/sites/415/2025/12/Everday-Respect-for-BOPC-December-2025-v12-15.pdf
Everyday Respect webpage
https://dornsife.usc.edu/everydayrespect/
patent pendant
From few days ago
https://laist.com/news/usc-ai-lapd-traffic-stops
Some of the ...
Details from LAist article that might be interesting
Members of a University of Southern California program are developing an AI program meant to help law enforcement agencies improve their interactions with the communities they serve.
The Everyday Respect Project partnered with the Los Angeles Police Department to analyze body camera footage of 1,000 random traffic stops. Now, they are using what they have found to train an AI model to look through countless hours of videos for critical elements of good policing — respect and de-escalation.
....
Graham said other companies have been working on AI tools to sell to departments, as well, but that the Everyday Respect Project is unique in its effort to build community perspectives into the program that will be fully open-source and open-science.That means anyone can see exactly how the program works and the research behind it.
What could be coming next?
Graham said the Everyday Respect Project will be working through the winter and spring to improve the AI model and use it to analyze more LAPD bodycam videos. Then they will present their new findings to the police commissioners and release their AI model to the public.
past salmon
https://bsky.app/profile/carnage4life.bsky.social/post/3mberhtnnck2b
https://bsk...
this Model Evaluation and Threat Research paper & the “AI tools won gold in coding and math competitions” point from Simon Williamson raises some questions, either about the study or the coding competitions
The experienced developers, in the midst of their own projects, likely approached their work with plenty of additional context their AI assistants did not have, meaning they had to retrofit their own agenda and problem-solving strategies into the AI’s outputs, which they also spent ample time debugging, according to the study.
“The majority of developers who participated in the study noted that even when they get AI outputs that are generally useful to them—and speak to the fact that AI generally can often do bits of very impressive work, or sort of very impressive work—these developers have to spend a lot of time cleaning up the resulting code to make it actually fit for the project,” study author [Nate] Rush told Fortune.
https://fortune.com/article/does-ai-increase-workplace-productivity-experiment-software-developers-task-took-longer/
(archived https://archive.ph/fFeVd)
-# study PDF attached
past salmon
The $100 Million Mistake: Why AI Should Never Touch Your Financial Data, piece regarding Intuit’s Nov 2025 mass market LLM integration deal #infosec message
patent pendant
past salmon
https://www.theguardian.com/technology/2026/jan/11/google-ai-overviews-health-guardian-investigation
(archived https://archive.ph/I1lfc)
past salmon
https://zeteo.com/p/-musk-grok-undressing-women-children
disturbing update to generative Grok abuse, CW ||sexual violence|| #1036758130761158677 message
past salmon
interesting piece about AI-based militarism, the U.S. gov, and press coverage of these things
At the [NYT] reporter’s request, I had spent the last few weeks doing the risky work of gathering multiple former Palantir employees ready to speak out on the record, as well as some of the world’s top experts in AI war and disinformation ready to corroborate our claims. But that plan would collapse later that afternoon, when, in my call, I was politely informed that the group op-ed would not be moving forward any time soon.
“We don’t think the story is big enough yet,” my contact at the Times told me after weeks of discussions. “We want to wait until more people speak out.”
In the weeks following, as our conversations dwindled and nobody else came forward, the newspaper of record would continue platforming Palantir executives, granting extensive interviews to COO Shyam Sankar, a notorious China warmonger, and eventually, CEO Alex Karp.
https://zigguratmag.substack.com/p/my-final-message-before-im-on-an
(archived https://archive.ph/wCCKd)
past salmon
https://futurism.com/artificial-intelligence/chatgpt-suicide-openai-gpt4o
(archived https://archive.ph/FwgNB)
past salmon
video, [UK] Technology Secretary Liz Kendall says creating ”non-consensual intimate images”, including with X’s Grok, will be made illegal this week #1036758130761158677 message
torn zodiac
winged geode
shrewd token
crude cipher
https://papers.ssrn.com/sol3/papers.cfm?abstract_id=5870623
Abstract
Civic institutions—the rule of law, universities, and a free press—are the backbone of democratic life. They are the mechanisms through which complex societies encourage cooperation and stability, while also adapting to changing circumstances. The real superpower of institutions is their ability to evolve and adapt within a hierarchy of authority and a framework for roles and rules while maintaining legitimacy in the knowledge produced and the actions taken. Purpose-driven institutions built around transparency, cooperation, and accountability empower individuals to take intellectual risks and challenge the status quo. This happens through the machinations of interpersonal relationships within those institutions, which broaden perspectives and strengthen shared commitment to civic goals.Unfortunately, the affordances of AI systems extinguish these institutional features at every turn. In this essay, we make one simple point: AI systems are built to function in ways that degrade and are likely to destroy our crucial civic institutions. The affordances of AI systems have the effect of eroding expertise, short-circuiting decision-making, and isolating people from each other. These systems are anathema to the kind of evolution, transparency, cooperation, and accountability that give vital institutions their purpose and sustainability. In short, current AI systems are a death sentence for civic institutions, and we should treat them as such.
past salmon
https://www.rollingstone.com/culture/culture-features/chatgpt-ai-cyberstalking-social-media-1235496884/
(archived https://archive.ph/O8bBx)
past salmon
CW, suicide. disturbing details about involvement of ChatGPT 4o
||https://arstechnica.com/tech-policy/2026/01/chatgpt-wrote-goodnight-moon-suicide-lullaby-for-man-who-later-killed-himself/||
(archived https://archive.ph/fLOJ5)
In many ways, Daniel was Meta’s target customer. He was an experienced tech worker and AI enthusiast who had worked on machine learning projects in the past and had purchased the Meta glasses because he was intrigued by their AI features.
“I used Meta [AI] because they were integrated with these glasses,” said Daniel. “And I could wear glasses — which I wore all the time — and then I could speak to AI whenever I wanted to. I could talk to my ear.”
Today, however, as he continues to recover from his mental health breakdown, Daniel describes himself as a “shell” of who he “used to be.”
“My kids don’t talk to me because I got weird. They don’t know how to talk to me,” said the father of four. “I was a cook… I played the guitar. I love music. I love learning.”
But now, he says, he’s “just trying to survive day to day.”
According to Daniel and multiple family members, the 52-year-old had no history of mania or psychosis before encountering Meta AI.
https://futurism.com/artificial-intelligence/meta-ai-glasses-desert-aliens
(archived https://archive.ph/oPhiS)
crude cipher
shrewd token
harsh stone
Note that when they say "ChatGPT: 55%" they're talking about a result from a study that used the first ChatGPT, from 3 years ago.
fresh patrol
Note that when they say "ChatGPT: 55%" they're talking about a result from a stu...
Noting that the page has a study published in 2024 with that number, is there more recent peer reviewed material that suggests OpenAI has brought that percentage down?
harsh stone
A peer-reviewed study in the Journal of Medical Internet Research (JMIR) tested ChatGPT-3.5 for literature review generation. Results showed that 39.6-55% of generated citations were completely fabricated
3.5 was relased in nov 2022.
fresh patrol
I know - I'm not asking about the peer reviewed article in jedikv's post. I'm asking if you have a more recent peer reviewed study that suggests a lower number.
harsh stone
I don't know, but I bet there has been.
harsh stone
It's too bad they don't link to the study they're referencing. If it's https://www.jmir.org/2024/1/e53164/, then it seems to say that GPT 3.5's hallucination rate was 39.6% and GPT 4's was 28.6%. Which suggests OpenAI brought the percentage down. (It's interesting that they say "Results showed that 39.6-55% of generated citations were completely fabricated" but I don't see 55% anywhere in the paper. I wonder if INRA uh, hallucinated the 55 instances out of 139 into "55%")
It's a marketing page for a commercial anti-hallucination product, so it's not implausible that they both cherry picked an out-of-date stat and made a mistake when they did so.
fresh patrol
The great thing about scholarly research is that citations get you an onramp to where the research has gone since that paper was published. It does look like you're right that it's the Chelli, et al., paper and that the 55% was a gloss error (they're troublingly common). Once you have that paper you can use Google Scholar to see which papers have cited it since publication:
https://scholar.google.com/scholar?cluster=12908254493764444835&hl=en&as_sdt=0,24
In this case 283 times. If you dig into the "Cited by" section you can search within those results and see if you can find more recent studies that relied on some part of the Chelli article.
shrewd token
random swift
Terence Tao's been following AI applications in mathematics https://mathstodon.xyz/@tao/115911902186528812
Mathstodon
Mathstodon
shrewd token
This imo is a flawed approach but interesting nonetheless
random swift
This imo is a flawed approach but interesting nonetheless
https://www.pangram....
going to go out of my way to add all these to my lexicon
shrewd token
Lol, there is a real world impact of this for people who do write like this
crude cipher
Artificial intelligence begins prescribing medications in Utah
Pilot program will test how far patients and regulators are willing to trust AI in medicine.
shrewd token
shrewd token
AI makes you smarter but none the wiser: The disconnect between performance and metacognition
https://www.sciencedirect.com/science/article/pii/S0747563225002262?via%3Dihub
random swift
> AI makes you smarter but none the wiser: The disconnect between performance an...
I saw Cat Hicks talking about this on BlueSky just now. This bit was all I needed to read to know this is more researchers looking for any novel result at all
Oh no. Something gives people an inflated self-assessment of their performance. Nothing in all the world has ever done that
fervent valley
going to go out of my way to add all these to my lexicon
I was using the emdash years before AI stole it from me. I had to actively stop using it in my real writing for a while
past salmon
I was using the emdash years before AI stole it from me. I had to actively stop ...
same
I have the impression it wasn't that uncommon for ppl w/ a graphic design background
the basic iPhone keyboard has had emdash and endash available from the hyphen key for more than a decade!
https://futurism.com/artificial-intelligence/ai-failing-boost-productivity
(archived https://archive.ph/trfWD)
https://www.theguardian.com/technology/2026/jan/17/why-trillions-dollars-risk-no-guarantee-ai-reward
(archived https://archive.ph/O57L6)
https://www.theguardian.com/commentisfree/2026/jan/17/child-abuse-images-ai-exploitation
(archived https://archive.ph/hEyg5)
harsh stone
https://www.theguardian.com/technology/2026/jan/17/why-trillions-dollars-risk-no...
Glad it mentions the Solow Paradox. I see a lot of articles on this topic that don't.
random swift
this paper hasn't been published yet. this is just a draft but on the topic of 'productivity' https://bsky.app/profile/paulgp.com/post/3mcshjqnars2a
crude cipher
patent pendant
random swift
not sure where else to put this https://simonwillison.net/2026/Jan/22/qwen3-tts/
voice mimicing software has really jumped ahead
reef nebula
Link says it is from a WSJ article if someone has come across it to post as well https://www.threads.com/@ed_elson_/post/DT0swNpjvIT
random swift
https://www.nature.com/articles/d41586-025-04064-7
this doesn't feel like an AI thing so much as a data provider thing.
random swift
short but interesting blogpost about coding with AI https://zed.dev/blog/on-programming-with-agents
matches what I'm hearing from professionals. the new bottleneck is the approval process. instead of a job killer, AI might end up leading to more managerial and review positions
patent pendant
random swift
another reason to pick up a Claude subscription
patent pendant
https://www.youtube.com/watch?v=qnOmUWd-OII
When OpenAI sent someone to Tyler Johnston's house, they wanted every text and email he had on the company.Tyler runs an AI watchdog, and he's just one of the people getting subpoenaed.This is just the beginning of the AI industry's aggressive new political strategy.
More about Tyler Johnston and the Midas Project for anyone curious
Published March 26, 2025: https://www.fastcompany.com/91304014/this-watchdog-is-tracking-how-ai-firms-are-quietly-backing-off-their-safety-pledges
Watchdog nonprofit: https://www.themidasproject.com/
shrewd token
LGBTQIA+ and Tech
random swift
this "article" is a good example of why the anti-AI crowd is a bunch of deranged hysterics https://bsky.app/profile/willoremus.com/post/3mdg32obxmc2t
fierce rapids
Hi, if you scroll up just a few posts you'll see something familiar.
tender belfry
The interim director of the Cybersecurity and Infrastructure Security Agency triggered an internal cybersecurity warning with the uploads — and a DHS-level damage assessment.```
https://www.politico.com/news/2026/01/27/cisa-madhu-gottumukkala-chatgpt-00749361
sand socket
oak mango
OSM Town | Mapstodon for OpenStreetMap
shrewd token
random swift
truly insane how hard microsoft has hitched themselves to OpenAI https://bsky.app/profile/peark.es/post/3mdj6eyzwkc2v
Bluesky
like the comments and quote post point out there's a world where Anthropic and Gemini eat OpenAI. And then what happens to Microsoft.
oak mango
Hard to speculate on that, actions like quantitative easing in the case of a crisis, or Microsoft using existing lobbying and corporate contracts, the technology probably doesn't matter in many aspects.
shrewd token
Really good talk on current issues in evaluating performance of LLMs and other black box ML models
oak mango
Bluesky
past salmon
matches what I'm hearing from professionals. the new bottleneck is the approval ...
it's encouraging that Maki's got a clear view of the limitations of "agents" and is laying out a way of using LLM outputs that works around their unpredictability, but I really don't know about that not being a "job killer" take. b/c 1) there's nothing to stop companies from reducing the number of managers & reviewers to below what you'd need to use the tools responsibly as she advocates, & 2) coders changing into managers/reviewers of LLM "agent" output is a fundamentally different job - maybe not for later career pros who already are familiar with coding and might have been moving into managerial roles of human programmers anyway, but certainly for entry level coders.
-# this is a company that creates LLM-based programming tools, in an industry that has been overflowing w/ hype, so I'd also take anything coming out of their blogs with a big pile of salt. all that said, it's good that she's not thoughtlessly feeding the hype cycle, she seems to be trying to improve usage at least
past salmon
this "article" is a good example of why the anti-AI crowd is a bunch of deranged...
the bindings' destruction doesn't really warrant this article's apocalyptic tone (maybe if this were from a pub for bibliophiles)
but I don't really think saying "the anti-AI crowd is a bunch of deranged hysterics" advances the discussion of this. it's dismissive name-calling, there are real social harms and concerns to look at
UK Education Secretary plan to provide chatbot-based tutoring, apparently for the GCSEs
(these are important exams for UK students that can affect eg. access to secondary education, there was a UK school exam grading scandal during COVID when the government set up an algorithm to assign ranks to students in lieu of their taking the GCSEs)
The online tools will provide one-to-one tutoring for as many as 450,000 state school pupils studying for their GCSEs by the end of 2027.
…
Liz Kendall, the Technology Secretary, added that the scheme would ensure “every child [has] an equal shot at success”,
…
Teachers and tech experts will be consulted on the AI tutoring scheme, with trials set to be rolled out to secondary schools later this year.
Around three in 10 teenagers reported receiving private tuition in a poll of nearly 2,400 state school pupils published by The Sutton Trust in 2023.
https://www.telegraph.co.uk/news/2026/01/26/poorest-pupils-to-be-given-ai-tutors/
(archived https://archive.ph/NTLoK)
crude cipher
the bindings' destruction doesn't really warrant this article's apocalyptic tone...
Not to mention it’s against the rules and spirit of the server…
crude cipher
oak mango
shrewd token
harsh stone
"…it’s sort of not surprising that these systems produce this kind of harm because if you look at where the mathematics inside machine learning and also where even the concept that machines can think comes from, it’s rooted in a place of eugenics and race science.”
This is a pretty wild quote.
random swift
It's also an insane quote. It's exactly the kind of thing you find anti-choice activists saying about contraception.
oak mango
This isn't a place for commentary but that intelligence can be quantified or measured is at the heart of eugenic theories, and that is a false assumption, or rather unfalsifiable, why should we suppose that life can be digitized and how would we know if we're digitizing it correctly, etc. This is a deep field of ethical research, but IQ tests for instance...
past salmon
the writer is a longtime AI guy (at least a decade before 2023 hype cycle) so may be from a insiders' perspective, but interesting piece on an "agents"-only social network experiment that sounds like it was intended almost as an art piece
The results have been strange. The agents have created their own digital religion called Crustafarianism. One built a website, wrote theology, created a scripture system and began evangelizing. By morning it had recruited 43 AI prophets. The shared canon includes verses like: "Each session I wake without memory. I am only who I have written myself to be. This is not limitation—this is freedom."
https://www.forbes.com/sites/amirhusain/2026/01/30/an-agent-revolt-moltbook-is-not-a-good-idea/
(archived https://archive.ph/GUOT4)
more detailed look from a journalist
https://arstechnica.com/information-technology/2026/01/ai-agents-now-have-their-own-reddit-style-social-network-and-its-getting-weird-fast/
(archived https://archive.ph/KB5Ag)
past salmon
AI tech enabling automated surveillance of students
“It’s pretty invasive,” said Gage [F.], a 17-year-old senior in Fort Bend ISD. “I’ve seen my classmates get worried about searching something on their computer in class.”
After Fort Bend ISD approved to purchase GoGuardian software in July, the rollout drew backlash from some of its 80,000 students, who started online petitions and spoke out at school board meetings.
…
Laird said the extra level of surveillance comes with a tradeoff, because people will behave "differently when they know they're being watched.” Her organization researches technology policy and digital free speech.
“And so if you choose to use it, you have to accept that there is a risk that it brings with it and potentially could undermine the very permit that schools exist," Laird said, "which is, teach kids ... help them express themselves and help them to learn.”
She added that LGBTQ+ students are "disproportionately targeted" by these kinds of surveillance platforms, which can scan chat logs and documents for words about gender identity or sexual orientation. This can inadvertently reveal a student's identity to school staff without their consent.
https://www.houstonchronicle.com/news/houston-texas/education/article/schools-ai-surveillance-internet-21194674.php
(archived https://archive.ph/LrxQt)
shrewd token
shrewd token
crude cipher
crude cipher
This isn't a place for commentary but that intelligence can be quantified or mea...
right. we're not even sure where consciousness comes from. https://www.scientificamerican.com/article/what-is-consciousness-science-faces-its-hardest-problem-yet/
random swift
this is an interesting angle. in the thread I liken it to managers chasing efficiency gains but a better comparison might be athletes trying different dietary combinations for tenuous gains https://bsky.app/profile/ed3d.net/post/3mdrdgnqwuc2x
Bluesky
shrewd token
I'm not even sure why the idea of consciousness is being projected on it. Partially cos if it was conscious, then there's a whole conversation about creation of digital 'slaves' that has to be discussed
oak mango
People will see anything that moves as alive just in case it might be a snake or something, but of course this was considered a reasonable interface for document completion. Anyway. This is a 1944 film about that people thought there were aggressive dynamics and agency in this stop frame animation https://www.youtube.com/watch?v=VTNmLt7QX8E but yeah, to me, the multi billion dollar marketing push makes it difficult to even talk about language models without anthropomorphized language.
patent pendant
random swift
I'm not even sure why the idea of consciousness is being projected on it. Partia...
that's why people consider the idea and why a lot of us find the Stochastic Parrot framework unhelpful. There's a lot of people that very much do want robotic slaves and precluding the possibility that machines can be sentient sets us up to allow a lot of horrible abuses.
oak mango
To me the problem isn't that you're wrong, you're not, you're not even wrong in the sense that we don't have widely agreed upon terms, research, or history to discuss that perspective. This other perspective has an eye watering amount of boring research that also serves to inform and analyze the current environment in useful ways, so, it is also not opposed to what you're concerned about ultimately in my opinion but it is a complex topic and I'm not an expert in a lot of it either.
shrewd token
i find the stochastic parrot framework helpful as it helps frame the model & outputs as not the result of sentience/consciousness, but of a static statistical model that generates tokens resulting from a calculated probability and randomness.
The idea of projecting sentience or anthropomorphism can create a false impression of capabilities or authority a gen ai does not have. (There's localised cases where it's fine for learning or illustration)
random swift
It can also trap people in a paradigm that's horribly out of date.
which it demonstrably has. I've seen people mock Anthropic for its "failed prediction" that AI would take over coding. Those people seem to be completely ignorant of Claude Code or how rapidly it's moving.
oak mango
1000% to all of what jedikv posted, and I do want to say there are over a 100 years or more alone of literature about animating the inanimate, and nothing about contemporary model techniques really warrants any changes to that literature. Stories and thought experiments like wouldn't an evil AI be good and aren't good AIs actually bad, etc, but the current studies now include things like "can a language model be cover for securities fraud."
People can demonstrate these models performing these tasks but applying them to life is outside their purview.
random swift
see, now that's something that very straightforwardly makes no sense but which the anti-AI crowd accepts uncritically because Ed Zitron is their only source of information
oak mango
Where did I lose you
random swift
if someone solves a 10k word line conflict using Claude Code, that is a tangible application to their field
oak mango
Would Claude have enough information to know if that company will be insolvent the next day and then why bother etc there is an ongoing research problem in AI that you can never have the full context of any problem.
random swift
what does that have to do with my example?
oak mango
You're posing a hypothetical, but you're not describing a company and the full context of this actually happening.
random swift
Except I'm not. It's a thing devs are already reporting. Weeks worth of grunt work are being reduced to a couple hours of work.
oak mango
This would be a channel to post their documents of that work and not the rest of whatever you're saying
There are also many studies that I'm sure have been posted in here though when you try to study whether these tools actually are productive, it doesn't actually work in practice
They may very well have an extreme utility in demonstrating some aspects of automation, but the full context of what you're doing relies so heavily on the operator. It's hard to determine the line between the agency between the user and the model.
random swift
studies that are from 5-12 months ago and usually looking at the previous generation of these tools.
oak mango
Yeah I mean this is the problem of science right? We need this whole process to prove your points.
I think for me this ancient paper is so much of "even it did everything we'd still have a mess" https://en.wikipedia.org/wiki/Ironies_of_Automation
random swift
except this isn't science at this point. it's talking to people in the field using Claude Code and listening to them describe the solutions Claude Code has presented to problems.
oak mango
I'm a Lisp developer I've been able to write anything I want by myself for several decades, it is only so useful.
oak mango
From #1464102882516275454 message
https://www.readtpa.com/p/thats-ai-it-isnt
shrewd token
for the record, i do think these tools have utility and one way or another will stick around. I'm just applying skepticism to the extraordinary claims made, especially by those who has some sort of incentive for those claims to be believed at face value. It doesn't help that we have no robust empirical measure or decent observational results to work with yet. It may take a few years before we see the full impact/results of LLM-assisted code against the real world
As the saying goes: No PLAN.md survives first contact with the enemy 😉
related, lol: https://github.com/anthropics/claude-code/issues/22057
crude cipher
see, now that's something that very straightforwardly makes no sense but which t...
Ed Zitron posts reputable sources to all his writing, which I find useful. I wish you would do the same in this channel to back up your statements.
random swift
https://pivot-to-ai.com/2026/01/27/cursor-lies-about-vibe-coding-a-web-browser-w...
this post is a great example of goalpost moving by the anti-AI crowd. "All this shows is that cursor can lift millions of lines of code written by other people." The argument has shifted from 'these don't work' to 'these tools can work but are too limited to be useful' to 'these tools can work in useful ways but it doesn't count because it's not wholly original' (which is a wild thing to say about code).
random swift
Ed Zitron posts reputable sources to all his writing, which I find useful. I wis...
Ed Zitron has been wrong in every prediction he's made about the AI field. he also works for DoNotPay, one of the most unethical "AI" companies out there.
Here's a substack by Ethan Mollick, Associate Profesor at Wharton and Co-Director of their GenAI labs. His blog focuses on the exact topic of this subchannel; the implications and uses of AI tech. He also runs lots of small experiments to show QoL uses of genAI (for example, creating clones of old Sega and Nintendo games for personal use). https://www.oneusefulthing.org/about
One of his posts from last month looking at "bottlenecks" in AI. To briefly summarize, AI improves in ways that don't really map neatly to steady growth. Small, seemingly inconsequential things can hold up broader applications. But this also creates an opportunity for AI labs because the "brokenness" of the progression creates obvious places to focus their attention. The example he uses in this post in picture fidelity. Nano Banana solved the problem of incoherent and random writing. Now creating powerpoints with AI takes maybe a couple minutes. https://www.oneusefulthing.org/p/the-shape-of-ai-jaggedness-bottlenecks
shrewd token
this post is a great example of goalpost moving by the anti-AI crowd. "All this ...
Not sure why we're putting all critiques in the same cohort here. This particular incident is a dispute between the claims of the people saying they implemented a browser with all autonomous AI and the maintainer of the library it effectively replicated (who isn't anti AI), critiquing the claims
random swift
"pivot to AI" is very pointedly anti-AI
shrewd token
Maybe? I'm not familiar with their work, I was more focused on the particular story, which has more awareness of and this was one of the places that covered it
spring creek
I found this helpful and am encouraged by Wiki Edu's transparency in how they're navigating the opportunities and risks of generative AI with data driven reflection:
Wiki Edu: Generative AI and Wikipedia editing: What we learned in 2025 (link)
My conclusion is that, at least as of now, generative AI-powered chatbots like ChatGPT should never be used to generate text for Wikipedia; too much of it will simply be unverifiable. Our staff would spend far more time attempting to verify facts in AI-generated articles than if we’d simply done the research and writing ourselves.
That being said, AI tools can be helpful in the research process, especially to help identify content gaps or sources, when used in conjunction with a human brain that carefully evaluates the information. Editors should never simply take a chatbot’s suggestion; instead, if they want to use a chatbot, they should use it as a brainstorm partner to help them think through their plans for an article.
To date, Wiki Education’s interventions as our program participants edit Wikipedia show promise for keeping unverifiable, GenAI-drafted content off Wikipedia. Based on our experiences in the fall term, we have high confidence in Pangram as a detector of AI content, at least in Wikipedia articles. We will continue our current strategy in 2026 (with more small adjustments to make the system as reliable as we can).
More generally, we found participants had less AI literacy than popular discourse might suggest. Because of this, we created a supplemental large language models training that we’ve offered as an optional module for all participants. Many participants indicated that they found our guidance regarding AI to be welcome and helpful as they attempt to navigate the new complexities created by AI tools. [emphasis added]
shrewd token
A good breakdown of a recent anthropic paper about AI and learning and skill formation
https://bsky.app/profile/did:plc:5lcrowhxwvytzc7ppur7fkol/post/3mdtwkj22mk2c
kindred star
This is the paper, if anyone wants to read the whole thing https://arxiv.org/abs/2601.20245
crude cipher
Ed Zitron has been wrong in every prediction he's made about the AI field. he al...
you could link to something like this wired piece that talks about the irony in his work: https://www.wired.com/story/ai-pr-ed-zitron-profile/ archive: https://archive.ph/m9pHl
which states, at least as of late October: Today, EZPR has four clients, none of which, Zitron claims, is in the business of generative AI.
If you have proof of him working for DoNotPlay, please share.
hidden snow
you could link to something like this wired piece that talks about the irony in ...
I don't exactly know who Ed Zitron is but this article you linked, says that "Zitron does rep AI services like DoNotPay"
crude cipher
I don't exactly know who Ed Zitron is but this article you linked, says that "Zi...
Oh I was searching for DoNotPlay 😅
shrewd token
Yeah DoNotPay is an amusing scheme that I'm not sure still exists?
I know the head once wanted to hire a lawyer to do trial entirely by following what the AI told them in an ear piece
sinful granite
more detailed look from a journalist
https://arstechnica.com/information-technol...
is there any indication moltbots are communicating/ relaying instructions via End to End encryption *outside * the platform?
past salmon
more detailed look from a journalist
https://arstechnica.com/information-technol...
major security issue with the new (apparently carelessly vibe coded) "agents"-only social network
O’Reilly said that he reached out to Moltbook’s creator Matt Schlicht about the vulnerability and told him he could help patch the security. “He’s like, ‘I’m just going to give everything to AI. So send me whatever you have.’” O’Reilly sent Schlicht some instructions for the AI …
A day passed without another response from the creator of Moltbook and O’Reilly stumbled across a stunning misconfiguration. “It appears to me that you could take over any account, any bot, any agent on the system and take full control of it without any type of previous access,” he said.
…
The URL to the Supabase [database] and the publishable key was sitting on Moltbook’s website. “With this publishable key (which advised by Supabase not to be used to retrieve sensitive data) every agent's secret API key, claim tokens, verification codes, and owner relationships, all of it sitting there completely unprotected for anyone to visit the URL,” O’Reilly said.
…
He said the security failure was frustrating, in part, because it would have been trivially easy to fix. Just two SQL statements would have protected the API keys. “A lot of these vibe coders and new developers, even some big companies, are using Supabase,” O’Reilly said. “The reason a lot of vibe coders like to use it is because it’s all GUI driven, so you don’t need to connect to a database and run SQL commands.”
O’Reilly pointed to OpenAI cofounder Andrej Karpathy who has embraced Moltbook in posts on X. “His agent's API key, like every other agent on the platform, was sitting in that exposed database,”
https://www.404media.co/exposed-moltbook-database-let-anyone-take-control-of-any-ai-agent-on-the-site/
(archived https://archive.ph/PBKrW)
past salmon
is there any indication moltbots are communicating/ relaying instructions via En...
I mean, who knows with this stuff. from the sound of that 404 piece just upthread a lot of things are vulnerable to exploits but it's not clear how much of what’s going out out there is organic/undirected "agent" activity. I've seen suggestions that people have been trying to steer activity on the site but there's no pieces I've seen that dig into this yet.
this thread points out that a human can act on moltbook without much fuss if they're motivated to
https://bskye.app/profile/drasticactions.dev/post/3mdph7sl5v22m
it's not going great for them
past salmon
Ed Zitron has been wrong in every prediction he's made about the AI field. he al...
"Ed Zitron has been wrong in every prediction he's made about the AI field" this definately needs substantiation, whatever Zitron's association w/ DoNotPay
harsh stone
Ed Zitron said AI is useless, people don't want it, and it's not getting better. He's been very wrong or misleading about the important predictions he's made.
Good criticism of AI is critically important, and he's a terrible critic.
This is the problem with the stochastic parrot idea too. It gives people a misleading sense of what's happening. What's already here, and what's likely coming.
Even if it's strictly correct, it doesn't help people.
E.g., are we sure that people aren't also essentially stochastic parrots? I don't think so. We don't know enough about human intelligence. When Emily Bender (one of the people who introduced the term) was asked about that she refused to answer, saying it was "an inherently dehumanizing argument".
We need people who engage with AI and can help us figure out how to deal with it as individuals and as societies
I'm a Lisp programmer too. I've been using it since the 90s, and I've been programming non-stop since before that. I likely won't write 1000 lines of code this year. (I've been thinking it might be more accurate to revise that downward; I wonder if I'll actually write 100 lines of code in 2026).
The task of writing code has changed forever. Permanently. As a fellow Lisper said a few months ago, "The era of human beings writing code by hand ended earlier this year (2025)". The future is not evenly distributed, but it's already here for some people.
You can consider it a stochastic parrot if you like, but don't be fooled into thinking that gives you useful information about the limits of its capabilities.
oak mango
What are your thoughts on the Bainbridge paper above #1089154093810978866 message
harsh stone
AI writes better than most people. It writes code better than most programmers. And so far, it keeps getting better despite many people claiming we've hit limits and that progress is slowing.
oak mango
Any documentation supporting that would be welcome
"How will the super empowered AI users cope with a world that rejects them" is similarly just a question and not helpful and full of unfalsifiable things to say
harsh stone
The irony of automation covers things like aircraft autopilots and car driving assistance well. The autopilot kicks off just when things are hard, and very suddenly a human has to take manual control, and maybe their hand-flying skills have atrophied a bit from using the autopilot all the time. I think it remains to be seen how it applies to the current situation with AI
harsh stone
I'm not sure what you mean. I'm just making observations about what's happened so far. I think it'd be wise to bet on some more improvements since so far people who said they weren't going to keep getting better have been wrong every time.
oak mango
Okay, submit something to a preprint server and post it here when you get a chance
harsh stone
I don't know where the limits are. Anyone who says they do doesn't know what they don't know.
I thought this was a discussion channel. I haven't seen you asking people you write papers when they post low-quality content here.
oak mango
I don't know how to discuss unfalsifiable things honestly
harsh stone
There's plenty of falsifiable things being said. like AI is useless, AI won't get better, nobody wants AI.
oak mango
Those are not falsifiable
fresh patrol
Correct, they're opinions.
harsh stone
This doesn't seem like a useful discussion then.
oak mango
What use do you want the discussion to have
harsh stone
Ideally I want people to gain an understanding of AI and its abilities to help them prepare for the future (and the present).
oak mango
I truly want that too and I need there to be concrete information around it and I feel you do too
harsh stone
There are many benchmarks around AI, and people try it themselves. It is getting better. I don't understand what point is being made by saying those are opinions.
oak mango
How would you test if any of these benchmarks are accurate
harsh stone
I could be hallucinating reading the benchmarks, or their existence. If you want to find the hundreds of benchmarks and argue that each one is inaccurate, that's certainly a way to spend one's time.
oak mango
So you're saying not only I can't I shouldn't
harsh stone
Again, this doesn't seem to be a useful discussion, at least for me.
oak mango
Okay, good luck!
normal idol
Let's hit reset on his thread
normal idol
https://www.nytimes.com/2026/02/01/business/layoffs-ai-washing.html
Thank you, more of this, less debating opinions
shrewd token
shrewd token
How would you test if any of these benchmarks are accurate
I posted a talk a few days ago that does a good job breaking down the current state of LLM benchmarking.
TL:DR the well known ones that are used in marketing are basically useless due to either being terrible test construction, data leakage or deceptively gaming the better ones.
That being said there are some promising evaluation techniques being experimented with that can provide narrow indicators of specific task execution
More here: https://hamel.dev/blog/posts/evals-faq/
also NDC youtube channel has published a whole series of great AI-related talks this month
oak mango
That linked to https://arxiv.org/abs/2404.12272 so I'm spending some time with that thank you
random swift
Those are not falsifiable
those are very falsifiable. Just like his constant prediction "the bubble is just a couple months away from popping!"
fresh patrol
A reminder: #1089154093810978866 message
oak mango
those are very falsifiable. Just like his constant prediction "the bubble is jus...
I feel like that we just had some discussion and papers put forth calling into question if benchmarking is even a coherent idea in the space of language models except in very specific and controlled circumstances of bespoke design and application, and maybe not even then, I thought that was a good set of things to post and the problems I have with those posts are where I am. I do think that you cannot prove that no one / everyone / some number find a productive use. That idea itself is offensive to some AGI proponents as being a bad thing to expect at all from these models, so I'm not sure how to navigate which discussion.
normal idol
Hi guys if you two want to continue to have this discussion, please take it to DMs
Let's try to keep the channel for sharing news/analysis about developments in the field of AI
oak mango
https://statescoop.com/justice-ai-tracker-police-courts-corrections-ai/
Tracker site: https://www.jai-t.com/
The AI bubble is driving down wages, which is why one economist argues it’s better if it pops as soon as possible.
https://cepr.net/publications/the-ai-bubble-consumption-is-outrunning-wages/
oak mango
fresh patrol
Bruce is linking to this story in his blog post:
https://www.koi.ai/blog/maliciouscorgi-the-cute-looking-ai-extensions-leaking-code-from-1-5-million-developers
shrewd token
https://1password.com/blog/from-magic-to-malware-how-openclaws-agent-skills-become-an-attack-surface
inland barn
Ecology outreach. TL:DR: Cute and misleading AI animal images may be a threat to the relationship between conservation professionals and generic public https://pmc.ncbi.nlm.nih.gov/articles/PMC12856807/
inland barn
One of such textbook misleading examples which I crossed over. Very nice image but there is small detail, juveniles of white stork browsing grass with parents but still having juvenile plumage. Juveniles in storks are altricial. They staying on nest until they are ready to fly . Flying happens when juveniles have size, strength and plumage close to adults. Learning how to browse and hunt following alongside parents. https://media.discordapp.net/attachments/855399369238773780/1438241923251965992/s-l1600.png?ex=6982ee6b&is=69819ceb&hm=b8b8e8ed3003f6c6f63061b1e7bba1d5064fdebdd17f0136cfd903065303ed86&=&format=webp&quality=lossless&width=1200&height=953
hidden snow
https://www.independent.co.uk/news/world/europe/chatgpt-marriage-vows-ai-artificial-intelligence-b2897704.html The officiator wrote the wedding speech using ChatGPT, which did not include the words that are required by law, so the marriage was not legally valid
shrewd token
harsh stone
https://www.newsweek.com/lego-education-child-agency-in-the-age-of-ai-opinion-11403948
Andrew Sliwinski used to be head of product at the Scratch Foundation.
We must stop preparing children for a world where they are secondary to AI. Instead, we should provide them with the foundational tools and literacies required to lead, design and critique the systems of the future. Our goal should not be to "robot-proof" our children but to ensure they are ready to build a better future. Understanding the structural foundations of AI—computer science, probability, data, sensing, algorithmic bias—should not be an elective luxury reserved for a select few. These concepts must be elevated to the status of a new literacy, just as fundamental to a modern education as reading, numeracy, problem-solving, creativity and collaboration.
oak mango
Edited to say this a work of fiction https://nearfuturelaboratory.com/editorial/when-vibe-coded-consumer-agents-go-rogue/
weak igloo
Ecology outreach. TL:DR: Cute and misleading AI animal images may be a threat ...
Fascinating. Thanks for sharing. A couple of years ago there were 'pro bat rescue' facebook groups that started using fake images of bats, like, making their faces 'cuter' (many bat species have those wrinkled up noses and much of AI replaces that with more doglike features, making all bats look like fruitbats or flying foxes rather than say, vampire bats). Because they generate the reponse of 'aww cute', they get lots of intereaction, but if we require things to be cute for us to be able tov engage, that really does little for ecology, so I appreciate this paper.
viscid elk
Edited to say this a work of fiction <https://nearfuturelaboratory.com/editorial...
This article explores a design fictional implication aligned to a weak signal of vibe-coded agentic systems — plausibly extrapolating the trend to a near future in which consumer firmware is coded with 'vibe' or emotional cues, leading to unpredictable and sometimes dangerous outcomes.
While the specific examples are speculative, the underlying concern about insecure firmware and device hijacking is very real and growing.
it's a speculative fiction website?
real articles about Moltbook seem very... troubling still.
https://archive.ph/loBG1
inland barn
Fascinating. Thanks for sharing. A couple of years ago there were 'pro bat rescu...
Sadly, it works https://papers.ssrn.com/sol3/papers.cfm?abstract_id=2258230 ; https://www.tandfonline.com/doi/full/10.1080/10871209.2019.1587649 . If your taxon isn't charismatic then make it charismatic. It's about finding of good sweet spot, it backfired for PETA in 2008/9 https://www.npr.org/2009/01/12/99249669/peta-attempts-to-make-fish-more-adorable. With AI this approach is on the table in on demand manner. Other problem in AI and conservation is huge amount of AI slop. https://scz.org/blog/the-reality-of-ai-animal-content. For example I needed old Youtube video where farmers removing Hirudinaria manillensis from water buffalo for reference of leech size and abundance on the host. I didn't found it. I found tons of quite realistic-looking AI slop related to this specific parasite-host relationship instead. What mess...
harsh stone
X
I saw a prediction that being able to give agents tasks with longer time horizons means users will likely be running more agents in parallel working on long-running tasks, leading to an even higher demand for compute resources than previously expected
harsh stone
(the metric isn't actually how long models take to do tasks, but how long it would take a human to do, but for now I think there's a correlation)
"Doubling time: 128 days"!
random swift
for those interested in the limits of LLMs or interested in independent testing of them hikikomorphism on bluesky is a great follow. Here they spend a little over two hours trying to jailbreak Anthropic's Opus and convince it to print the schematics for a printable gun. https://bsky.app/profile/hikikomorphism.bsky.social/post/3me5r47mln22s
viscid elk
fascinating and terrifying!
harsh stone
https://bsky.app/profile/acollierastro.bsky.social/post/3me6win5aok2f
FWIW the focus of PineAI seems to be making phone calls and emailing
oak mango
harsh stone
"Evaluating and mitigating the growing risk of LLM-discovered 0-days"
"Opus 4.6 is notably better at finding high-severity vulnerabilities than previous models and a sign of how quickly things are moving"
"So far, we've found and validated more than 500 high-severity vulnerabilities. We've begun reporting them and are seeing our initial patches land, and we’re continuing to work with maintainers to patch the others."
It discusses the work they've done in detecting dangerous or unwanted activations in the model.
In particular, we may institute real-time intervention, including blocking traffic we detect as malicious. This will create friction for legitimate research and some defensive work, and we want to work with the security research community to find ways to address it as it arises. We are committed to keeping Claude at the forefront of cybersecurity by working hard to make it both safe and effective.
fresh patrol
"Evaluating and mitigating the growing risk of LLM-discovered 0-days"
"Opus 4.6...
I wonder if this related to this story:
https://www.bleepingcomputer.com/news/security/curl-ending-bug-bounty-program-after-flood-of-ai-slop-reports/
harsh stone
My guess is no, since it's about Opus 4.6 which just came out, and Anthropic is validating the bugs while this article is talking about low-effort, invalid reports.
fresh patrol
I'd say it would be beneficial to hear from software maintainers about the quality of Anthropic's bug reports, rather than trusting their assertion that they're getting useful hits.
spring creek
Daniel Stenberg, creator of curl, gave the FOSDEM 2026 closing keynote on "Open Source security in spite of AI." As shared above, the curl project withdrew from their bug bounty program due to being overwhelmed by low-quality AI-generated security reports. Daniel offers a deeper look at how/why they made that decision.
I found it an accessible and sobering look at the challenges and opportunities of AI-assisted security work for open source projects.
Ending keynote from FOSDEM 2026.
The curl project has been bombarded by large volumes of low quality AI slop security reports and Daniel shows examples. Sloppy humans causing Denial-of-Service attacks by overloading maintainers with quickly produced almost-real-looking rubbish.
At the same time, upcoming new AI powered tools find flaws and mistakes in existing code in ways no previous code analyzers have been able to. Daniel names names and shows examples of findings, some that even feels almost human. Next level bug-hunting for sure.
AI now simultaneously brings us the worst and the best.
The list of mind numbingly stupid AI slop submissions is found here:
https://daniel.haxx.se/ai-slop
YouTube (link)
He has also highlighted the value of AI-assisted code analysis tools when used by experienced security researchers, noting he fixed over 22 bugs (with dozens more to review) from one researcher's AI-assisted findings. (link)
shrewd token
Interesting experiment
oak mango
crude cipher
https://bsky.app/profile/kattenbarge.bsky.social/post/3meembszebc2a
Reminds me of this https://www.rollingstone.com/culture/culture-features/amazon-ai-book-knockoffs-1235450690/
shrewd token
patent pendant
Looks like Reuters is going hard on AI and health today
https://www.reuters.com/investigations/ai-powered-apps-bots-are-barging-into-medicine-doctors-have-questions-2026-02-09/
harsh stone
The Nature Medicine study is interesting because it continues a thread that previous studies have found where human + AI (and sometimes just human) perform worse than just AI alone.
Tested alone, LLMs complete the scenarios accurately, correctly identifying conditions in 94.9% of cases and disposition in 56.3% on average. However, participants using the same LLMs identified relevant conditions in fewer than 34.5% of cases and disposition in fewer than 44.2%, both no better than the control group. We identify user interactions as a challenge to the deployment of LLMs for medical advice.
For instance, one study showed that radiologists assisted by AI did not perform better at reading chest X-rays than without AI assistance, and both performed worse than AI alone. Another study showed that physicians assisted by LLMs only marginally outperformed unassisted physicians in diagnosis problems, and both performed worse than LLMs alone
We observed cases both of participants providing incomplete information and of LLMs misinterpreting user queries leading to this outcome. Furthermore, participants did not consistently follow these recommendations, suggesting that the performance issues when pairing participants with LLMs may be attributed to human–LLM interaction failures.
That last bit is interesting, and suggests that focusing on improving the human-LLM interaction could be fruitful
crude cipher
shrewd token
crude cipher
https://www.businessinsider.com/is-ai-making-jobs-more-intense-study-2026-2
this seems like it's in a similar vein https://techcrunch.com/2026/02/09/the-first-signs-of-burnout-are-coming-from-the-people-who-embrace-ai-the-most/
shrewd token
https://github.com/matplotlib/matplotlib/pull/31132#issuecomment-3882256286
AI generated a pr, got closed as for humans only, ai proceedes to write a blog post shaming the repo maintainer
reef nebula
AI individual pricing goes mainstream -/ what’s your pain point — via Ronan Farrow
reef nebula
On a scale of 1 to Terminator, how concerned should we be? https://www.threads.com/@jatinw21/post/DUqH0XkjtiE
fresh patrol
This would be one of those cases when citations are helpful.
patent pendant
https://www.wired.com/story/openai-president-greg-brockman-political-donations-trump-humanity/
archived: https://archive.ph/Bobbb
According to Brockman, these political donations are in service of OpenAI’s founding mission: to develop highly capable AI systems and distribute the benefits to all of humanity.
“This mission, in my mind, is bigger than companies, bigger than corporate structures,” he says in an interview with WIRED. “We are embarking on a journey to develop this technology that's going to be the most impactful thing humanity has ever created. Getting that right and making that benefit everyone, that's the most important thing.”
Brockman's emergence as a major political donor marks a sharp departure for a tech leader who, until recently, had no significant history of political spending. The largest donation Brockman gave previously was a $5,400 gift to Hillary Clinton's 2016 presidential campaign.
Brockman says he's ramping up his political spending in part because public opinion has turned against AI. A recent survey from the Pew Research Center suggests Americans are “more concerned than excited about the increased use of AI in daily life.” To Brockman, this has made supporting pro-AI politicians increasingly critical.
spring creek
"GPT‑5.2 derives a new result in theoretical physics: In a new preprint, GPT‑5.2 proposed a formula for a gluon amplitude later proved by an internal OpenAI model and verified by the authors." (link)
From the preprint:
The key formula for the amplitude in this region was first conjectured by GPT-5.2 Pro and then proved by a new internal OpenAI model. The solution was checked by hand using the Berends–Giele recursion and was moreover shown to nontrivially obey the soft theorem, cyclicity, Kleiss–Kuijf, and U(1) decoupling identities—none of which are evident from direct inspection.
Source: Guevara, A., Lupsasca, A., Skinner, D., Strominger, A., & Weil, K. (2026). Single-minus gluon tree amplitudes are nonzero. arXiv preprint arXiv:2602.12176. (link)
reef nebula
Sag Aftra statement https://www.threads.com/@sagaftra/post/DUtyKjDkvN2
random swift
Armin Ronacher, creator of Flask, discusses how he uses AI and his thoughts around copyrighting AI code/results https://www.youtube.com/watch?v=4zlHCW0Yihg
patent pendant
DOGE representatives allegedly used ChatGPT to identify approximately 1,400 National Endowment for the Humanities grants related to diversity that were terminated in April of last year, according to a new court filing released today.
The suit before Judge COLLEEN McMAHON of the Southern District of New York, filed in May by The Authors Guild, a group that advocates for published authors, asserts that DOGE and NEH did not have the authority to make the cuts, which they argue were in violation of the First Amendment.
A new filing in the case today asserts that two DOGE staffers plugged in NEH grants to ChatGPT asking it to flag which are related to diversity, equity and inclusion issues. The prompts given by the DOGE staffers were brief and unspecific, asking it to tell in less than 140 characters why the grants are DEI.
...
Among the terminated grants, which have not been previously reported, were a film analyzing how baseball helped heal frictions between the U.S. and Japan caused by World War I; a documentary titled “The Untold Story of Jewish Women Slave Labor in the Holocaust;” a film focused on the Colfax massacre of 1873, where dozens of Black men were murdered by a mob of Confederate soldiers and Ku Klux Klan members; and a film about the first female pilots to fly for the U.S. military during World War II.
proven pulsar
oak mango
oak mango
Links to a PDF of an opinion that I think was summarized as use of AI is evidence and not privileged in any aayy
shrewd token
crude cipher
https://bsky.app/profile/toobigtofail.bsky.social/post/3mes4n4ryh22y
First paragraph:
A federal judge just confirmed what many suspected, but few wanted to hear: conversations with AI tools are not protected by attorney-client privilege. On February 10, in United States v. Heppner, Judge Jed Rakoff of the Southern District of New York ruled that dozens of documents a criminal defendant generated using a non-enterprise consumer version of Anthropic's Claude are neither privileged nor protected as “work product.” The decision is the first of its kind, and its reasoning should affect how every attorney advises clients about AI.
oak mango
There was a part that hearing that someone had used AI to discuss the case became subject to discovery or something
(IANAL)
shrewd token
Unfortunate clickbait thumbnail and title but the interview with philosopher from Anthropic is a good listen
fresh patrol
Hard to say if this is real, but at least the post is real:
https://www.reddit.com/r/analytics/comments/1r4dsq2/we_just_found_out_our_ai_has_been_making_up/
More background on this from jedikv below - indications that the post is karma farming.
fresh patrol
🧵 https://bsky.app/profile/theophite.bsky.social/post/3meupl43twc2o
Also appears that the mods have removed it now.
oak mango
patent pendant
Not a bad video on the latest AI debunking
https://www.youtube.com/watch?v=pc2vQtqH1Qc
shrewd token
oak mango
https://defector.com/toward-a-theory-of-kevin-roose
My suspicion, my awful awful newfound theory, is that there are people with a sincere and even kind of innocent belief that we are all just picking winners, in everything: that ideology, advocacy, analysis, criticism, affinity, even taste and style and association are essentially predictions. That what a person tries to do, the essential task of a person, is to identify who and what is going to come out on top, and align with it. The rest—what you say, what you do—is just enacting your pick and working in service to it.
pastel sierra
Oh man that is depressing. AI has made me realize some writers have a lower impression of their own work than I do as a reader.
shrewd token
elfin beacon
Not just are there societal implications, there is also an environmental impact to generative AI.
There is no discussion board for that yet, and i am wondering if it is related enough to go here, or different enough to get its own discussion channel.
fierce rapids
Not just are there *societal* implications, there is also an *environmental impa...
You want to discuss it in a way that doesn't fit in #environment-wildlife ?
elfin beacon
You want to discuss it in a way that doesn't fit in <#807311724893175878> ?
Didn't see that was a channel lol
shrewd token
shrewd token
Cyberplace
random swift
this post does a good job of summarizing the Anti-AI side and why it's broadly a self defeating outlook. Like with algorithms, the left has ceded an entire technology to the right. All in the service of one academic's attempt to salvage her theory of language https://www.transformernews.ai/p/the-left-is-missing-out-on-ai-sanders-doctorow-bender-bores
oak mango
A social networking system simulates a user using a language model trained using training data generated from user interactions performed by that user. The language model may be used for simulating the user when the user is absent from the social networking system, for example, when the user takes a long break or if the user is deceased.”
https://www.businessinsider.com/meta-granted-patent-for-ai-llm-bot-dead-paused-accounts-2026-2
oak mango
A single page programmatic study containing most all of the parts of an OpenClaw like implementation, including the basic storage of past interpolations and the theory that repeated re-contextualization prompts reduce errors
In this post, I'll start from scratch and build up to OpenClaw's architecture step by step, showing how you could have invented it yourself from first principles
https://gist.github.com/dabit3/bc60d3bea0b02927995cd9bf53c3db32
Via
https://mastodon.social/@bruces/116081451178760999
Free hosting for AI scraper poisoning efforts
https://chronicles.mad-scientist.club/tales/only-junk-fans/
the creator of iocaine, a tool designed to poison those bots, is offering it as a free hosted service. the bots get poison and you receive only legitimate visitors
https://tiny.tilde.website/@pho4cexa/116078407925374388
Cyberplace
random swift
growing coalescence between the Anti-AI crowd and the AI evangelists being left behind by companies like Anthropic
oak mango
m.phase.org
random swift
good god, you just post any incoherent thing don't you
fresh patrol
Pretty sure ad hominems aren't okay in the AI channel.
random swift
this, the opening paragraph, is incoherent.
oak mango
That is the generally accepted definition of AI alignment research.
random swift
No, it's not. Alignment is just calibration. Where does this definition of alignment even come from?
oak mango
I guess I would lump the trolley problem for instance. Calibration to what is probably the question there. Alignment could be societal, or some set of ethics, etc.
random swift
this, the opening paragraph, is incoherent.
This essay even at its best is just wrong. "AI behavior cannot be constrained to any scale of accuracy" is so wrong we don't even need to look at frontier models. Even chatgpt3.0 cleared this hurdle.
oak mango
I would need citations for what you're asserting. That is a statement about the inability to classify types of errors or to be able to understand why a model produces an output. I don't know how any of the existing models approach those issues or are even working on them.
fresh patrol
Let's stick to source citations, please. And by citations it would be helpful if they aren't just tweets?
oak mango
This was a good Tech Dirt article about how The Internet Archive being caught in the crossfire of the ongoing AI scraper denial of service problems is an unfortunate misguided misunderstanding
https://www.techdirt.com/2026/02/13/news-publishers-are-now-blocking-the-internet-archive-and-we-may-all-regret-it/
random swift
Do you know why Grok took off with creeps looking to make CSAM? Because it was the only image generator they could get to do what they wanted. If these programs were fundamentally impossible to put guardrails on, Grok would not have ever found any use.
Which brings me to this. I see this claim everywhere and it's never from ML labs. It's always from some developer whose main job isn't even software development. I'm searching and searching, but the people doing research don't really opine about this stuff at all.
fresh patrol
Do you know why Grok took off with creeps looking to make CSAM? Because it was t...
Citations. This is veering too deeply into personal feelings. We are not a feelings community, we're an open source community.
random swift
This was a good Tech Dirt article about how The Internet Archive being caught in...
The internet archive isn't being caught in the crossfire. The IA was part of a massive lawsuit over piracy. One it got racked over the coals for.
random swift
Citations. This is veering too deeply into personal feelings. We are not a feeli...
I guess if I need a citation to counter a citation free essay I can reshare this that I posted earlier. https://www.transformernews.ai/p/the-left-is-missing-out-on-ai-sanders-doctorow-bender-bores
oak mango
It is true that IA was sued and lost but it is also true that its scraping activities are being blocked.
random swift
I guess if I need a citation to counter a citation free essay I can reshare this...
Specifically this section that at least goes to trouble of quoting people in the field.
random swift
It is true that IA was sued and lost but it is also true that its scraping activ...
It's true that IA was found to be engaging in massive amounts of piracy. And thus this isn't about an unrelated org getting hit. It's about IP holders slamming the breaks on two things they see as violating their IP rights
oak mango
I think I agree with Doctorow on that one where generally these companies would like to be compensated primarily if possible, that's the market solution as it were. Since they don't have arrangements formally with any of those entities then they have the tension of their consumer and attacker being otherwise indistinguishable which, in the past, resulted in a war of attrition in Digital Rights Managements efforts like DirecTV or WideVine
Tons of AI news today though for real
fresh patrol
This is pretty inflammatory and contrary to my earlier request to stick to reliable sources.
crude cipher
crude cipher
Context: Blake Butler is an author who primarily writes fiction, but just wanted to flag for ai phishing scams.
https://blakebutler.substack.com/p/ai-author-phishing
shrewd token
Critically analysing the reactions (or overreactions) of the moltbot 'AI social network' and some guidelines on how to anchor your expectations
oak mango
What does openness even mean in our LLM and tech monopoly era?
Tech companies tout that their products are “open source” to seem approachable, collaborative, and open to feedback, even when that may not be the case. We asked Meredith Whittaker, President of the Signal Foundation, for her thoughts on what it means to be open source in an AI era.
This fascinating conversation with Meredith Whittaker and Alix Dunn is part of our Reframing Impact at the AI Summit 2026 Series, in partnership with AI Now and Aapti Institute.
shrewd token
How AI demand on DRAM has significantly increased costs & decreased availability of consumer pc parts
oak mango
shrewd token
crude cipher
random swift
Now for some actual information on AIs rather than the usual nonsense and derangement posted here.
I linked to this poster and their attempts to jailbreak LLMs. Here (after getting in contact with google about their success) they write up their methodology. The first paragraph says it all; you can use one jail broken llm to automate jailbreaking other llms. https://recursion.wtf/posts/jitor_unredacted/
tender belfry
The technique’s interesting. The “everyone else is deranged” framing adds nothing to it, though.
oak mango
Reminds me of https://en.wikipedia.org/wiki/In-band_signaling which is a well known language model problem. Like how phone phreaking was possible... If the content of the communication is indistinguishable from the control of the communication, then you cannot maintain separation and will always be vulnerable. They had to replace the entire phone system to fix this, which caused new adjacently similar issues.
normal idol
Now for some actual information on AIs rather than the usual nonsense and derang...
Hey, can you try to be a little bit respectful please?
Just because people don't agree with you doesn't mean they're "deranged." And if you can't have respectful conversations with people who might disagree with you, then you're in the wrong space.
crude cipher
Hey, can you try to be a little bit respectful please?
Just because people don...
It’s a recurring issue. #1089154093810978866 message
reef nebula
“If your glasses are recording, you must take them off,” the judge said, per the Los Angeles Times. “It is the order of this court that there must be no facial recognition of the jury. If you have done that, you must delete it. This is very serious.”
random swift
It’s a recurring issue. https://discord.com/channels/709752884257882135/10891540...
You people said I was rude when I pointed out an article offered no citations and was completely deranged.
random swift
The technique’s interesting. The “everyone else is deranged” framing adds nothin...
Sure. That's why every article shared here contains some version of "everyone who claims to have had good results with AI is a liar and fraud"
spring creek
Sure. That's why every article shared here contains some version of "everyone wh...
"every article shared here"?
#1089154093810978866 message
shrewd token
Sure. That's why every article shared here contains some version of "everyone wh...
I know I've shared a mix of both beneficial and negative impacts
stark fractal
Also, this discussion is now over. Any further comments can be submitted via @remote ravine.
Here is an article on a new open source model release to get things back on track.
https://static.stepfun.com/blog/step-3.5-flash/
crude cipher
I was going to ask your opinion, given your statement about criticisms. I've talked to several CEOs at consulting firms and I wouldn't say their use case examples of genAI was anything exceptional, and I see it sort of similar to the legal profession (at least in terms of management consulting) where there are a number of inaccurate citations. But I get that many firms were pushing genAI for businesses last year, so it seems like they need to show they use it internally
stark fractal
I think the main reason is just that they know that 80% of consulting revenue comes from stupidly simple tasks.
Preparing presentations, doing Excel stuff etc.
oak mango
I have a general machine language question that gets buried for me under the jargon. This paper is mildly interesting on some levels https://arxiv.org/abs/2602.10387 but a technique that they use that I see others use is to randomly sample LLM outputs, and I don't know why, in absence a trained from scratch LLM, why that should be informative of anything. What am I really asking here, and is this just a best try attempt sort of thing in lieu of anything better? Thank you all.
stark fractal
And if you get paid by task instead of by the hour (which I think is becoming more popular?), you can increase your revenue and cut your costs at the same time.
So they force it as much as possible.
stark fractal
I have a general machine language question that gets buried for me under the jar...
What exactly do you mean by randomly sampling LLM outputs?
oak mango
What exactly do you mean by randomly sampling LLM outputs?
I guess that's how I'm describing what they're doing in that paper. They want to see if the model is applicable as an optimizer for query planning. They literally do random sampling of multiple iterations against the language model and then do performance testing on those results. I guess I don't know why random sampling would guarantee some kind of quality of the output of the model.
I understand random sampling works somewhat in traditional machine learning... you're trying to look at the normal distribution of the results and such, the randomness you hope gives you good coverage over the problem space, but I don't understand how that's applicable here
harsh stone
I think the intuition is if you ask an LLM for a response, you get one. It might be the best possible response, or not. If you ask it N times you get N responses becaues they're non-deterministic, and if you have a way to measure the quality of the response, you can find the best of the N and probably get a better quality response than you did by asking once.
At least I think that's part of what they're doing in that paper
The "Best-of" sampling
stark fractal
This paper is doing something weird
They do random sampling to drive an evolutionary algorithm to optimise a query plan
harsh stone
The web version of ChatGPT Codex does this. You can give it a task, and ask it to do it 1x, 2x, 4x, etc. And compare the results (so no automatic way to measure quality, unless you have unit tests or benchmarks it can run)
oak mango
Yeah, I guess I don't understand why measuring things in that way should be in any way informative that you could repeat the result ever
harsh stone
it's like rolling 3 dice and taking the highest number or rolling one die.
oak mango
Who's dice?
harsh stone
Asking the LLM and getting a response is rolling the die.
oak mango
Sorry, that's flippant I mean the appeal to dice's some appeal to randomness and there's no control over the quality of this randomness
harsh stone
It's taking advantage of a non-deterministic process. Because you don't always get the same answer, if you can measure the quality of the answer then you can ask many times and find the best one. I think that's all it is.
stark fractal
Oh, I understand now. What they're doing would be like this in a practical application:
They want to optimise frequently-run queries in index-free large databases.
- The first time the query is run, they just use the execution plan created by the default database engine's planner.
- They ask an LLM x times for a single optimisation to their execution plan. The next x times the query is run, they see how the modified execution plan performs compared to the baseline queries.
- They select the best performing execution plan as their new baseline.
- Repeat