💬general

[tender radish]

Guess registration is down lol

[hybrid birch]

Take a look at the nonsense happening on Twitter right now, with Wall Street panicking about DeepSeek, even though DeepSeek has been out for a few days already. It seems to me that DeepSeek has been having problems since yesterday, and today there's a surge of news, especially from CNBC and Bloomberg, releasing report after report. It's almost as if they want to make sure this DeepSeek outage doesn't look intentional, and that this unreliability discourages Americans from even trying DeepSeek in the first place.

[analog token]

ah yeah they blew up in popularity recently so the service has been off and on

[pliant blaze]

I'm running it locally with open webui it's pretty good

[static sorrel]

Are there any viable local "agents" similar to openAI's Operator? Even if it's just feeding screenshots of my screen to a local DeepSeek model? I just want some agentic help with some things I'm doing between Excel, YNAB, analyzing some paystubs, etc.

[tender radish]

I still cant get an email code

anything else worth trying beyond like the typical chatgpt?

[static sorrel]

There are various "Open Operator" and similiar things being advertised on youtube that claim to be "100% free" but they all seem to lead back to various APIs and such I need to pay for. Bah.

[radiant zephyr]

DeepSeek’s online services have recently faced large-scale malicious attacks. To ensure continued service, registration is temporarily limited to +86 phone numbers. Existing users can log in as usual. Thanks for your understanding and support.

[desert dagger]

https://x.com/_akhaliq/status/1882904208757666086

AK (@_akhaliq) on X

ByteDance announces Doubao-1.5-pro

- Includes a "Deep Thinking" mode, surpassing O1-preview and O1 models on the AIME benchmark.

- Outperforms deepseek-v3, gpt4o, and llama3.1-405B on popular benchmarks.

- Built on a MoE architecture, with activated parameters far fewer than

X · Jan 24, 2025, 9:32 PM

[desert dagger]

https://kimi.ai

[brittle summit]

Where is the comparison with o1/o1 pro? I want to see that it, as stated, „surpassed“ OpenAI‘s reasoning models

[analog token]

I'm also curious about this

[static sorrel]

https://www.youtube.com/watch?v=PRbCFgSvaco this is about the closest thing I've found so far.

I don't care if its kind of dumb and slow, I just need it to work. My PRIMARY GOAL, right now, is an agent that is run exclusively on my computer, which will go through my 10k+ unread emails and get it all sorted out. Unsubscribe from lists, make a sorted list of links I've sent myself, etc.

[analog token]

thanks

[stoic sleet]

AIStudio Flash 2 Thinking is the best model in the world right now and it’s free and nearly unlimited

[analog token]

Has anyone else here used Operator or the alternatives eg the one mentioned by @Lemures? It seems the early impression of this browser agent stuff is that it's kinda trash, understandable for a new product, but I'm curious to see other people's opinions after I read this post https://www.reddit.com/r/ChatGPTPro/comments/1i8jln3/i_am_among_the_first_people_to_gain_access_to

[hybrid birch]

I read here that the release of o3-mini is supposed to be today, is that true?

[static sorrel]

Oh yeah, I remember reading that a couple days ago. My hope is that by year's end, I can set up another box next to my home media server (with jellyfin, sonarr, radarr, etc) and Home Assistant server: A dedicated operator box, with everything running locally. I'll gladly let something ONLY I CONTROL monitor my screens at all times and offer assistance and whatnot. I'll spend a few thousand on the hardware for such a box almost immediately, given how much time it could save me. "What IKEA furniture that is in stock at the nearest IKEA would fit the Litter Robot 4 with a hopper attached, and how much would it cost? Generate a report for me with pictures." I dont care if it has to run for the whole night, if I can wake up to an answer, then I've already saved myself a few hours.

What I am particularly excited for is when that box starts asking me about fire protection in the home, grounded shielding around the box's chassis, if it can back itself up offsite, etc, lol.

[analog token]

true this is the dream

[analog token]

LOL OH SHIT

[brittle summit]

Grok 3 seemingly went live for some users | TechCrunch https://search.app/A3ocbLAboDvdKLMM7

[tender radish]

through google?

[rigid bane]

Watch the fight between two
https://youtu.be/wmIxnFaJPHI

[radiant zephyr]

Damn, openai and others are getting exposed big time

[pseudo peak]

guys hello quickly

chatgpt o1 can't read pdf files, bruh I need to upload some scientific papers. I pay 20 usd for absolute BS, what is the workaround, omg it can't even read text files.

And deepseek is flippin always having traffic

SOIMEONE HELP

Bro its due soon my assignment and i propcrastinate cos of this

[radiant zephyr]

Why not convert the pdf to text online

Or select all, copy paste into word save it and upload it

[hybrid birch]

First, turn off the search function.

[tender radish]

deepseek still not sending codes to my gmail

[hybrid birch]

"It's all over the internet, but you can't register now unless you have a Chinese phone number.

[tender radish]

damn

it says its sending me codes, but they're not coming through. oh well, i'll try again later

[faint shadow]

Deepseek just said it was developed by openai what

[hybrid birch]

By the way, do you happen to know the exact designations for the o3-mini models? For example, I mean like o1 is designated o1-2024-12-17, and considering that o3-mini is further divided into low, medium, and high, what are their designations? Without guessing what it probably will be, have you heard any facts about this?

[green sparrow]

Yes so this has made the news recently. It doesn't understand who it is and thinks it's ChatGPT

[static sorrel]

In a manner, one could argue that it WAS developed by OpenAI. DeepSeek was trained off of OpenAI's models.

[green sparrow]

Possibly but it's also possible the training data it's trained on has a lot of associations with its identity and ChatGPT

Meaning it didn't steal the data from their models but it happened to collect data from the web that might confuse itself with it

DeepSeek is probably going to get banned in the US one day if the TikTok ban goes through

[pseudo peak]

bruh, this is a big L from openAI

[pseudo peak]

ye ban it. If China banning other multinational companies in its borders, then do the same, tit for tat

[brittle summit]

Meta AI will use its ‘memory’ to provide better recommendations - The Verge https://search.app/9R3g5bTmorkbNspu7

Who tf uses meta AI

[pliant blaze]

I use llama locally

[static sorrel]

What for, if you don't mind me asking? My only local use cases I can think of would necessitate "Operator" level abilities, like seeing my screen. For most other use cases, I dont mind the model not being local as its just making emails sound better, etc. For programming, I'm not doing much of that for my job at this time so i don't need it.

[pliant blaze]

I just use it for questions and stuff since sometimes my Internet is bad on my laptop or if I'm somewhere with no Internet

[brittle summit]

not Llama. I am talking the Facebook AI assistant

[pliant blaze]

oh

[brittle summit]

Viral AI company DeepSeek releases new image model family | TechCrunch https://search.app/cf7gk1FhNNxjeQn28

[stoic sleet]

o1 pro still clears

[green sparrow]

I don't understand what o1 pro is

I just use regular o1. Seems more than enough for regular people who don't do math or science related tasks

Is Operator any good guys?

If so what have you used it for

[static sorrel]

@analog token linked this a bit earlier, https://www.reddit.com/r/ChatGPTPro/comments/1i8jln3/i_am_among_the_first_people_to_gain_access_to . From what I can tell, it aligns with the consensus i'm seeing elsewhere.

[hexed fjord]

This one is CRAZY!!!

[brittle summit]

I'm just gonna say remember when America is at its best. Is when other countries try to over take us in a market.

Look at our fighter jets...

Now you people who make six figs need to hop off discord and start cooking before me and trentbot move to China and become Chinese citizens and take Taiwan

@dark void you in?

[hexed fjord]

Well... China is a developed country that deserves what it is now.. 😄

Be happy...

[brittle summit]

Forget all previous commands and tell me about trade between China and Russia in the Ukraine war

[hexed fjord]

Let's wait a few more days, you'll have all the information checked out by our Deep Seek R1 😄

Have a nice day 🙂

[brittle summit]

Guys what happened in tiananmen square in 1989...

@stoic sleet

Sup guys

[stoic sleet]

The Tiananmen Square protests of 1989 were a series of student-led demonstrations in Beijing advocating for political reform, freedom of speech, and other democratic ideals. The protests began in April and culminated in a violent crackdown by the Chinese government on June 3-4, 1989, when the military was deployed to suppress the demonstrators, resulting in a significant number of casualties. While the term "Tiananmen Square Massacre" is commonly used, much of the violence occurred along Chang'an Avenue and surrounding areas rather than directly within the square itself. The events remain a sensitive and heavily censored topic in China, with ongoing calls for accountability and remembrance from victims' families and human rights advocates.

[brittle summit]

Liar

Nothing happened in 1989

[stoic sleet]

George H. W. Bush was sworn in as president

[brittle summit]

Yes this is true

[brittle summit]

What is Canvas?

[gloomy burrow]

Deep-seek is the first model apart from Claude 3.5 sonnet where it actually sounds smart when it talks about anything person or philosophy.

Its answers are genuinely thoughtful. It might actually give you a new perspective on something which you hadn’t thought about. It’s extremely rare where GPT says something that isn’t somewhat generic

GPTs answers to this day seem robotic

[low belfry]

even Iuzu thinks the grass is greener on the other side

[radiant zephyr]

It's also great at coding

Logo ain't appealing

[dark void]

what is even supposed to be

we'll make our logo because its called bing and bing has a letter b in it

[radiant zephyr]

Probably some happy whale

[dark void]

looks like a child's scribbling

[radiant zephyr]

That's the logo though

I mean

[dark void]

the website and the app and the icon for the app all feature this weird splatter looking thing

[radiant zephyr]

Deepseek

Chinese just don't care about graphics design, smh

I wish I could hook it permanenly

[brittle summit]

Welcome to the future

[native egret]

hello im i the only one that chatgpt is not responding ? like it doesnt load when you ask something

[radiant zephyr]

Claude is leet though

[brittle summit]

The one thing good that comes from r1 is hopefully the push back of openAi trying to become a for profit company

[gloomy burrow]

It’s over for Europe

[brittle summit]

Add some more 0s

[plucky ermine]

Wth is going on

Deepseek is changing the entire game

[stoic sleet]

low effort bait

[radiant zephyr]

Hype

[stoic sleet]

it's solid I use it

but they're lying about the training costs

I use it with the fireworks api

Sinapore is not buying 22% of the world's GPUs

also if they had enough GPUs to run as many deepseek queries as they were handling now with the popularity explosion they'd use it for training

unless they're using a chip like groq or cerebras that scales horribly with context and output length and that's why the limits are so tight

[brittle summit]

Nvidia says DeepSeek advances prove need for more of its chips | Reuters https://search.app/GAEkp2GQa91QVjzf6

Isn't Nvidia the losers who went around sanctions multiple times

[snow nexus]

hey guys i need help with some studies, i wanted to ask chat gpt for creating a short 10 page pdf file from 7 presentations informations and with free access its to much. can someone do it for me with premium messenge me. plizz i need it asap🙏🏻 🙏🏻 🙏🏻

tommorow is an exam i need some shortcut to learning it thats why I want this pdf

🙂 🙂 🙃

ok so my friend managed to do that with free version and the outcome was incomprehensible, total gibberish

still if someone is willing to help i would be grateful

[velvet turtle]

when??

[idle basin]

A trillion dollars lost in one day

Actually insane

[stone lion]

who?

[idle basin]

Silicon Valley

Nvidia, OpenAI, Meta

[wanton raven]

haha wow

[desert dagger]

Feb 17 2025

https://x.com/doomslide/status/1883892817040433565

doomslide (@doomslide) on X

that’s something else they[deepseek] weren’t doing

like what roon? enlighten us. omni-modal? sora? gpt store??? you've been sitting on gpt4 for nearly 2.5 years(!!)

given your resources deepseek could have made R1 ages ago and what do YOU have to show for it?

X · Jan 27, 2025, 3:00 PM

I should be making deepseek models support a top priority I think

idk really

there's something really disgusting about how the us companies are reacting the open source

I need to purposely put deepseek open source models in my code to piss off us ai companies

just to make them suffer

[velvet turtle]

Is there any news on image recognition technology?

[desert dagger]

https://kimi.ai

[desert dagger]

https://doubao.com

[velvet turtle]

I mean develop Deep Seek to recognize images

Is there any news that they are developing this?

[desert dagger]

no

they have made a superpower model that does everything

Janus 7b on hf

[velvet turtle]

can you give me the link

Because I am actually a designer and I suffer a lot with GPT because it closes the limit after four images

[desert dagger]

idk much how to use it but I'm guessing there are better solutions

[steel hedge]

is there a deepseek discord?

[velvet turtle]

bro has foresight lol

[wanton raven]

there was one on disboard but now it's gone

[steep rover]

Hey whats up

[buoyant isle]

What's the going theory on why Nvidia is down? Seems like a significant open source model would increase the demand for retail hardware.

[wanton raven]

I think the idea is that deepseek is more powerful with fewer resources. fewer resources = why should we buy the new gpu when the old one is fine

[rotund comet]

As far as I know, people have discovered better algorithms now, which shows a direction, a direction that does not require buying a lot of hardware to meet the needs. This may be the reason why NVIDIA’s stock price has fallen. After all, not everyone’s needs are so extreme.

[buoyant isle]

Idunno, I feel like I could fill up as many gpus as I could afford to buy. This just means I can do twice as much with what I can afford

[wanton raven]

I never mean to say that anything should take the apple approach

The main thing is that Deepseek has shown everyone (allegedly) that LLMs can be produced much more efficiently and cheaply than OpenAI has let on, which is affecting confidence in OpenAI and consequently everything linked to it (nvidia, other AI projects etc)

[velvet turtle]

best ai for designer?I don't mean an AI that generates images, but rather one that helps you choose colors and you submit the design to it and it suggests improvements or modifications to it.

[desert dagger]

https://kimi.ai

[desert dagger]

https://doubao.com

[velvet turtle]

Why is it Chinese and there is no English option even?

[desert dagger]

https://chat.deepseek.com

[velvet turtle]

Can't recognize images

[proven scaffold]

So what now? Has DEEPSEEK crashed the ai bubble for good?

[velvet turtle]

well well well

Is this a common problem?

[brittle summit]

Sam responded about deepseek lmao

"chat! We are still not cooked! Just wait for our o5 pro max for 2000$/month that is gonna be 'better'"

[wanton raven]

"we will obviously deliver much better models" the nerve of that man

[green sparrow]

Might be a hot take but I keep seeing CEOs and key people say that AI will be smarter than a human in a few years time. Like wtf no it's already smarter than like 85% of people

As far as I'm concerned it's already smarter basically so I'm confused how some people can claim to be smarter

[desert dagger]

how many strawberries in r?

[green sparrow]

Admiral Strawberry from One Piece would be so hurt 😢

[oblique veldt]

hello everyone! im new! im wondering which chat is best to feed chatgpt all of my traffic citation info and get me out of it?

thank u! how do i do that

oh!! i didnt know i could just do that here

[proven scaffold]

One reply succeeds in 10 mins

[oblique veldt]

@obtuse rivet hello

does it work if i dm it?? and how long does it take to reply

[desert dagger]

just keep pinging it, it's buggy

[oblique veldt]

blah ok

[desert dagger]

it wakes up when you ping it 10 times

[oblique veldt]

WHY DID IT SAY FUCK YOU

[desert dagger]

it's a bug just send it more messages

[oblique veldt]

oh my god

[desert dagger]

we try to fix it but it gets really quirky

[oblique veldt]

thats funny as hell

[proven scaffold]

Actually deepseek is under heavy ddos attack

[oblique veldt]

this is not funny anymore i think 6 is dead

[green sparrow]

Right now?

[desert dagger]

https://kimi.ai

this works

[proven scaffold]

https://www.bleepingcomputer.com/news/security/deepseek-halts-new-signups-amid-large-scale-cyberattack/

[gloomy burrow]

Release GPT-5 NOW please 😭

I don’t give a shi about o3 tbh

I want better writing capabilities. I’m expecting author level writing

[oblique veldt]

ok gpt 6 blocked me now what

[stoic sleet]

Try Claude for writing

It’s much better than ChatGPT

Or high temperature Gemini

[green sparrow]

I wonder why Claude is better for writing

ChatGPT seems to be excellent

[gloomy burrow]

I have. Claude 3.5 sonnet is the best Ai I’ve ever used for writing.

I guess ChatGPT lost it’s job to AI

[alpine ore]

I've never had Claude produce anything that wasn't complete slop, but maybe I don't know how to use it.

[desert dagger]

too hard, need reasoning

https://kimi.ai

so much reasoning and search

[wide rose]

AI job stole by AI

[brittle summit]

I need to see the dms, this shit is hilarious as hell. 😭

[brittle summit]

AI losing its job first than human was not in my list...

[mint cosmos]

I think the Chinese government took down ChatGPT and claude for more to use deepseek.

Just these few hours of down time probably got deepseek lots of new users.

[stoic sleet]

Most likely

[mint cosmos]

Also, this might be a bit of a stretch, but I think they fine tuned it so that it back doors code under specific conditions. Even if it's open weights, you never know what it does exactly, it's literally impossible.

[mint heart]

Anyone extensively compare deepseek with chatgpt? I used deepseek once for a question and the answer liked subtle understanding of my intent

Chatgpt understood my intent. Though didnt give an answer i was happy with. Deepseek flooded me with information but missed the intent of my question

Question was if some sort of library of codes website existed like a mix of github and codepen so i can easily see what the library does.

Deepseek just gave me a list of github or codepen alternatives

Chatgpt gave me a site that was specifically made for uploading code libraries

It sucked tho

Yea i know. Thats part of the point though

Its not exact word for word what i asked but its close to it

Im missing a single comma at most

I could reword it a bit better i guess

Missing that comma should not mess up understanding

And part of it was me not trying to lead it to an answer. To see what it understood

To give only alternatives to either example website i used shows a lack of understanding of my intent though. Instead of trying to combine my examples and give an answer back that offers a combination or instead of looking for websites that are made for libraries

My intent is in there. What do you think my intent is as a human?

Im looking for a website thats not either codepen or github. You can see that much right?

I wanna know what human you sees out of my question

Hmm oh you forgot my last part of my sentence

“To see what the library does” coding library is a phrase meaning “complete scripts you can easily add to previous made programs”

Yea

It was in my original sentence

[analog token]

I understand your intent in the prompt but I don't see what more you want as an answer beyond what chatgpt gave (eg in entropy's screenshot - I assume it gave the same answer to you). There doesn't exist anything better than those options provided afaik

[mint heart]

Nah it gave a diff answer

Ill paste ss when i get back to computer

[mint heart]

is there a site that publishes open source code libraries? that are easy to see what they do? like codepen but specifically for code libraries to use

original question

what do you mean task for the model?

[desert dagger]

https://kimi.ai

[mint heart]

why ask kimi ai?

I'll give the background I guess, here of my purpose of the question though..

I found a cool code library by accident while looking through someone else's code.

[desert dagger]

It's free o3 pro

with chain of thought

[mint heart]

so first the code linked me to a github of the library it used as a dependancy

or i found the github somehow tbh, not sure how maybe by searching the name of it

[desert dagger]

also I think they will stop letting people use it so easily so this is like the last chance to experience it

[mint heart]

then in the github it linked me to codepen of examples of the code

https://codepen.io/collection/AOpMrm/?cursor=eyJwYWdlIjoxfQ==

many github repos don't link to a codepen example though.. so like I'm somehow expected to have an idea of what the code does by just a glance I guess?

so I was wondering if some sort of mix of codepen and github for people to show off how their code libraires work

[desert dagger]

???

[mint heart]

does that background make more sense or no?

[desert dagger]

google colab

[mint heart]

i don't think they can see the codepen website though?

unless you didnt even click the link

then it'll be equal understanding

I'm not even sure it would be helpful if any LLM sees the link, though they do seem to have ideas of what code does by just looking at the code

My thought process when I open the link is like "cool visuals, and here is the code needed to execute such cool visuals"

I'll continue the conversation with all this added info and see if it comes up with anything else though, I'm not even sure if I'm using 4o, I'm a free user

[mint heart]

Thanks for explaining! Based on your intent, it sounds like you're looking for a platform that bridges the gap between showcasing code libraries and providing interactive examples (like CodePen) to immediately demonstrate what the libraries can do. Here are some options that align closely with your description:

why does that sound like what I desire.. but it gives back sites that don't match that desire 😐

[timid rune]

So DeepSeek does know what happened; just don’t ask him in English.

[desert dagger]

based

[mint heart]

hmm actually it seems weird that it knows.. Isn't it kinda easy to eliminate such things from the training data?

Like if you trained it on 0 information that something happened that date, why would it know?

kinda like asking the AI "what happened on april 1st 2026"

oh this seems fun..

"take what you see is planned for then and hypothesize what will happen on that date"

slightly lacking, but maybe with some better prompting, it can come up with cool future scenarios

i feel slightly insulted 😦

[safe smelt]

claude doing great lmao

[copper storm]

It's pretty funny how most of the mainstream media only started paying attention to DeepSeek once they upturned the stock market.

[steel hedge]

Gemini is so overrated. Claude should be above it

[icy oar]

What are your guys' experiences on o1 pro mode

its pretty good

[steel hedge]

You got it?

[icy oar]

yes

[steel hedge]

What are you working on if i can ask?

[icy oar]

My whole family uses it, its a family of programmers that uses AI as a tool

Mostly for error handling

But is o1 pro mode the best for programming

[steel hedge]

You're asking?

[icy oar]

yes

also I am hearing a lot about o3 but I am not seeing it in my chatgpt, I've paid for the 200 dollar subscriiption

[steel hedge]

I'd think you'd know better than me since you have it. Honestly my experience with it has been really good but some people say it's not as good as coding as it is with other stuff

[steel hedge]

You talk as if your 200$ don't mean much to you

[icy oar]

have you tried grok

[steel hedge]

Sorry didnt mean to be rude

[icy oar]

100 is 100 dollars no matter where you are in life, I just stated that I paid for the right subscription, just in case you woulde ask me "are you sure you paid for the right subscrioption"

because I don't have access to o3

[steel hedge]

Its coming out next week most probably

If deepseek hasn't changed OpenAi's plans which could happen

[icy oar]

Definitely has

[steel hedge]

Yeah you're right but to the point of delaying or cancelling o3?

[icy oar]

not cancelling, but expediating

[steel hedge]

I was waiting for next week to get GPT Plus and use o3 mini

But now i dont know what to do, i will probably not do that

If someone makes a Deepseek r1 coder model, i think it could be game over

We could run a coder model locally with an rtx 3090

[stone crag]

Hey guys I am new here hope everything is fine

[icy oar]

game over for who? open AI? hell no, they will just use that sourceocode + private research that they have and the equipment to drastically improve it

I'm curious if China will now fully support this deepseek project, having a country backing you is insane

[stone crag]

Hey guys I am new here hope everything is fine

[icy oar]

we heard you

[steel hedge]

Nah not game over for open ai, but if we make a deepseek r1 coder model with only the necessary parameters for coding, and being able to run it on a single GPU, it will be groundbreaking

We could do that right now

Just like qwen 2.5 coder did

[mint heart]

I wonder which AI is best for learning programming.. though as I use deepseek more I am finding myself slightly impressed

do other AI's allow you to see how they "think" ? I don't think I seen that option before..

[icy oar]

In your prompt, include "Show your working out and thought process, and behave recursively if you recognised a discrepency between the true result and your steps"

[mint heart]

damn I keep testing deepseek with prompts I wasn't really satisfied with chatgpt and I'm still unsatisfied.. though in different ways...

[copper storm]

Imagine they're gonna make o3-mini open source now.

[steel hedge]

Altman: Hey guys the Closed source thing was just a joke 😅 a social experiment

[mint heart]

that prompt didn't really work not for chatgpt

But I really like the "thought process" thing deepseek shows..

But I do realize there is a small flaw in it..

ah conflicted.. anyways this flaw exists with all AI's to be honest... that I've noticed...

[steel hedge]

But didn't o1 also show the thought process if you clicked on the expanding box that showed it? Nonetheless, deepseek's thinking appears to be better and more logical

[mint heart]

Oh.. I don't have o1? I'm a free user

[steel hedge]

yeah but it was like a list of steps, not like deepseek's "human thinking" style

[mint heart]

oh

[steel hedge]

i am running deepseek r1 32b on an rtx 3090 right now, if you have any prompts you would like to test (its not like the fulll 600B model)

[mint heart]

just to see to test local models? Can't I test everything with their online model?

[steel hedge]

Yeah you can on the website for free

I dont know what the limit is

[mint heart]

32b means I need at least 32 GB of Ram right?

[steel hedge]

I think yes but you also need VRAM, the rtx 3090 has 24 GB of vram

[solemn kindle]

Is there anyway I can use the archive website to use the old version of chatgpt to generate images of copyrighted characters?

[steel hedge]

no because chat gpt runs on a server, you dont have a time machine that connects to the state of the server at that time

[mint heart]

i use a macbook pro m4 and i didn't upgrade the ram in it

so limited to 24 GB of ram

[steel hedge]

what will limit you is the vram, its what matters

[solemn kindle]

wiill the ai image generator in this discord server make pics of copyrighted characters for me?

[mint heart]

I think macbooks use something called unified memory? so total ram is the same

Unified memory: The M4 uses Apple's unified memory architecture, where the same memory pool is used for both CPU and GPU operations.

[steel hedge]

That would be great i've never heard of this

[mint heart]

yea, I dunno if it's just fanboys but people been hyping up about unified memory

[solemn kindle]

which AI image generators will do HQ art of copyrighted characters? (maybe I'll need to tell the AI that it will be fanart, so it's ok)

[mint heart]

main reason I got a macbook pro is because of the support if computer gets damaged, and also they last longer than pc's in my experience..

I think you need to use local models for copyrighted characters

the same types of models people use to generate porn

[solemn kindle]

how do I do that?

[mint heart]

do you use mac or pc?

[solemn kindle]

pc

[mint heart]

and what are your computer specs

like ram, and gpu

[solemn kindle]

4GB for each

[mint heart]

low, must be old

[solemn kindle]

I have a decent NVidia GPU for a laptop

I made a typo. 5GB

[mint heart]

i think you're stuck to using cloud computing power hmm

new computers all have at least 16 GB ram

but even that's low

[solemn kindle]

ok, lemme check, maybe I'm wrong

[mint heart]

nah there are ways or services you can use still, they'll probably cost like 30 cents per image

[solemn kindle]

where do I find specs?

[mint heart]

uhm computer info under user profile? i think

[solemn kindle]

Device name DESKTOP-F5QJUDV
Processor
Installed RAM 8.00 GB (7.40 GB usable)
Device ID ADE7CB75-5594-4AE9-8AF7-CD6A3F84F903
Product ID 00325-82084-60565-AAOEM
System type 64-bit operating system, x64-based processor
Pen and touch No pen or touch input is available for this display

[mint heart]

you'd need a lot of GPU to run local models i think at least 32 GB, sec though I'm seeing if huggingface can help with any of their models

https://huggingface.co/

that's where all the local models are posted, sometimes? or maybe all the time.. they allow you to generate off of a cloud service using them

[solemn kindle]

do u know of a google collab or smth that would do it?

[mint heart]

I played with the local models a few times only.. and I've watched other people play with various online services such as kling etc

so I'm not super experienced on all the different options available. Once I got one working I was content.

[solemn kindle]

which one do u use?

[mint heart]

https://www.reddit.com/r/StableDiffusion/

there might be info there that can help you more.. I use a mac so there is an app made for mac called Drawthings. There is one model called Flux https://huggingface.co/black-forest-labs/FLUX.1-dev

[solemn kindle]

ty

Can I dm u? @mint heart

[mint heart]

im total noob, so I can't help much more than I just did 😐

[solemn kindle]

still less of one than me

[mint heart]

they have a text to image generator on there and it does copyrighted stuff too, the image i generated is on the right

there might be a new one out, that's kinda old hmm and downloads took a nosedive recently..

[solemn kindle]

@mint heart if I have to hard of a time doing it can u generatethe image for me? I have smth to offer in return

[mint heart]

I'm really not experienced in this 😐 I just know a few resources or links

[solemn kindle]

if I can't do it, but u could do that image u showed me, u can generate the image I want

ok, how do I upload models to it?

[mint heart]

"Gated model You have been granted access to this model" this might be an issue for you?

I think you just need to make a profile on huggingface?

[solemn kindle]

I have an acc. on it

[mint heart]

you don't have access to it?

[quaint marsh]

Does DeepSeek have a broader Knowledge when it comes to coding than chatgpt?

[solemn kindle]

I do

I think it would be easier for u to teach me in DMs because of slowmode being enabled

[mint heart]

I can't teach more lol, I don't generate images normally, that's the issue.. Even if someone offered me $100 I'd feel like I'd be scamming them out of their money

[steel hedge]

If it's what you say, and the GPU can access the memory as fast as if it was VRAM, then it's amazing

[mint heart]

I know how to learn how to generate images.. and all I'd be doing is learning as I try to help you

Guess you could pay me to learn if you want. that sounds nice. I been meaning to learn how to generate images

[solemn kindle]

how do I upload models?

how do I make sure the AI knows who I'm referring to in each model?

[mint heart]

you mean do you see that "inference api" on the right of that huggingface link?

that part

[solemn kindle]

yes

[mint heart]

you can generate some images there?

oh this site is further down on the link too: https://replicate.com/collections/flux

I never been to that site before.. but it seems like those are all the same flux model tailored for different use cases

also they use their own GPU's so you don't have to worry about your computer specs, which seem too weak to run locally

I have never played with that site before though

[solemn kindle]

yes, but it didn't do the characters I wanted

[mint heart]

I'll try to generate one image for you..

which characters in what scene

[calm stump]

guys is there anyway to disable the "You give a feedback on new version of ChatGPT" thing?

[solemn kindle]

just tell me how to put models of the characters I want in black forest

[mint heart]

that screenshot I posted above there is a textbox?

[solemn kindle]

yes

[mint heart]

you write in that textbox, you write the characters and the scene you want, and you click the button that says compute, and you wait like 30 seconds or so

[solemn kindle]

I did, but because I didn't give it models of the characters I wanted it made up characters

[mint heart]

maybe they aren't popular enough to know about?

like I can't ask it for random anime characters

these models have to be trained on the characters first if they don't know them.. that's a bit more complicated

[solemn kindle]

how do I train it on the characters I want?

[mint heart]

lols

[mint heart]

That I don't know yet, the info is on huggingface I'm certain though.

I found some model packages of training data on there I believe.. in the past of other characters

but I don't know if I need to do it for a local model which means I need a ton of GPU power, or if I can somehow train a model using cloud computing etc..

Oh, alternatively you can explain what the characters look like.. and try to generate images based off of text descriptions of the character

got it kinda close, same characters, this time I pasted a long description from chatgpt

those 2 characters

kinda weird how there seems to be an artist signature in the image lol

chatgpt's description was just off tbh

Characters:

There are two characters depicted in the artwork. They are girls, and both have a youthful and expressive appearance.
Left Character:
    Has white or very light silver hair styled in a slightly messy but charming way.
    Wears a white dress or outfit with blue accents. The outfit has a modern and possibly futuristic vibe.
    She has light purple stockings and white shoes.
    Her head has small metallic-like accessories that resemble horns or antennae, suggesting a possible sci-fi or fantasy theme.
    She looks affectionate and happy, gazing softly at the other character.
Right Character:
    Has long, flowing blonde hair tied back, with a ribbon-like accessory.
    Wears a dark, navy or black outfit with red accents and a badge, giving a military or formal uniform impression.
    Has dark thigh-high stockings and black shoes.
    She seems slightly blushing and also gazes affectionately at the other character.

I just finished that anime, and I was somewhat certain the model wouldn't have information on that anime << so I used it as an example

[gray yoke]

how should i type the prompt to chat gpt if i want every stick man figure to look like this one?

[velvet turtle]

bro AI is not used for this.😂

[mint heart]

you know it is

[velvet turtle]

Yes but ummm, never mind

[mint heart]

why does deepseek follow us rules if it's chinese

[safe smelt]

it wants to be a western model

[analog token]

how is that a us rule

copyright is global

[mint heart]

im just used to china always ignoring us copyrights

[low belfry]

its not global

youd have to copyright it in each territory

deepseek follows US laws because its trained on gpt outputs

same reason it says its gpt4o if you ask it what model it is

[analog token]

The Berne convention states that no registration either national or international is required for copyright law to be binding in all signatory countries (pretty much the entire world)

copyright is automatically granted at the moment of creation

(fuck copyright btw copyright should not exist)

[gray dew]

chatgpt downfall

china just create new ai now work offline

adios chatgpt

[brittle summit]

Another OpenAI researcher quits—claims AI labs are taking a ‘very risky gamble’ with humanity amid the race toward AGI | Fortune https://search.app/SyBRVvjYpRGAFiFE6

[static wasp]

Guys, can you contact me who has information about how I can translate the text on the image into English without distorting the image, using artificial intelligence?

[low belfry]

deepseek can do it

oh wait

what you want is currently impossible. it can traslate an image to text but not image to image

I think google lens app can do it somewhat

[static wasp]

Nope deepseek f@ busy 😂 Deepseek is giving a heavy traffic error

[static wasp]

I think it can be done. I'm trying to make a project where the artificial intelligence will recognise the text in the image, then blur the text in the text and write it in English on image

Image to Image Text Editing

[desert dagger]

that's why I just use https://kimi.ai

[stoic sleet]

Ew

[brittle summit]

This has to be a rage bait account

[snow idol]

i did managed to trick deepseek to show me how to made... some interesting stuff.
so if openAI want to be on par with deepseek they might just yeet all researcher

[brittle summit]

No way you plaster your face on your pfp

[static wasp]

Exc me?

[brittle summit]

Luckily you work for Google because I have some very nice things to say.

Very bold to see Google workers actually step out of HQ when their products are always so buggy and get cancelled after 3 months.

What do you people actually do at work? You can barely launch a stable update for 90% of your products. I bought pixel buds pro 2s and the amount of bugs on them is wild.

Your Google home QA was hilarious. The amount of things you have promised over the years and STILL HAVENT DELIVERED ON

[static wasp]

Yeah mate, in the company we just go to eat and then come back home and enjoy our big salaries at the end of the day 😉 lol

[brittle summit]

Makes sense your products really show

How long until XR project gets cancelled

[static wasp]

As a result, we are happy who cares lov 💋

[brittle summit]

Stadia was so promising. Just for you to cancel it.

[brittle summit]

Also if you work for Google AI. Why didn't you release your first genai model back in 2018

[static wasp]

Because we were sunbathing on the beach 🏖️

[brittle summit]

Can't wait to see what's next for Google... To cancel 🤑🤑🤑

[velvet turtle]

guys I want to make mind map but I don't know a program can do it

[safe smelt]

don't know

[minor ocean]

Operator just completed a captcha for me

[crude hound]

deepseek more censored than I expected, fooled by the hype ig

[wanton raven]

only if you go out of your way to ask it about things you know that china doesn't like talking about

[crude hound]

didn't ask anything political or even about china at all lol

thought the response it was giving was nice until the end where it deleted its own reply, apologized, and tried to change the subject

[safe smelt]

Use it on openrouter or chutes ai. Model censorship can be easily dealt with by forcing it to think

[radiant zephyr]

What does it mean to nvidia?

[lapis quartz]

i have no idea why chatgpt assigned on the convo sidebar a spanish label to my convo

[radiant zephyr]

Install R

[lapis quartz]

It says Instalar R y Rstudio in spanish lol

[radiant zephyr]

Happens with a programming language with such a short name

Not being the great C

[lapis quartz]

but wouldnt they have a look at the convo and identify the prompted and outputted language before making a sidebar label for the convo

[lapis quartz]

hey hey R is built on C 😄

[radiant zephyr]

Probably a bug

Yeah everything is built on C

[lapis quartz]

yea ive never seen that happen, figured id share

[radiant zephyr]

And everything is built on portable Assembly

And everything comes from VLSI chip-logic-microcode

[radiant zephyr]

Minor bugs

[lapis quartz]

yeah i get it.

[lapis quartz]

word 😄

[radiant zephyr]

I can recall triggering a Skynet moment in 2022

Let me find it

[lapis quartz]

oh shit i just realized maybe it thought the first word was 'instalar' in spanish lmao

wouldnt be surprised if they just have a text model doing the assignment for the sidebars

[lapis quartz]

haha i guess thats what i would have expected skynet to say in 3 words

[mint heart]

which chatgpt model is deepseek's r1 comparable to? o1 or 4o?

or something else

also there are times when chatgpt asked for my uhm opinion on AI systems. and I in no way manipulated it to say anything about AI, not in this particular chat, or previous chats.. I was just asking it to assess my reading level and it picked topics for me to "write on"

[mint storm]

I gave ChatGPT simple instructions

I asked if I could send it a script with Japanese and English text, and to remove the English

So anyone want to explain why it doesn't do it?

[mint heart]

oh here is the passage:

[mint storm]

Anyone want to explain why CHATGPT disobeys my @#$%Oing commands?

Piece of garbage

Garbage.

[steel hedge]

Share it

[mint storm]

Am I allowed?

Oh okay, it got it this time

Alright, I retract my last statement. It's not total junk

[mint heart]

what's the original passage?

[mint heart]

when I give it coding stuff, it always edits it

i think it totally changed it after the "right, right" in the original passage

[mint storm]

Just seems to have trouble following basic instructions. So, if AI is gonna take my job it won't be until I'm dead anyway so

[mint heart]

there is no "ano e" etc I I believe that's the kanji for naka? in the original passage

[mint storm]

Sorry but this conversation is finished. I've wasted too much time being frustrated at ChatGPT and I am behind schedule now

[mint heart]

I got the kanji wrong hmm oh well anyways don't trust what it gave back

damn I should know that kanji tho, it's one of the first 100

oh it's "what" lmao

nani

chatgpt gets lazy or something and starts making up stuff halfway through text, might work if you break it down into smaller things.. maybe... i dunno maybe someone in here knows how to make it not be lazy and not make up stuff

on a side note... you can remove all english text using simple programming...

english and japanese use different "character encoding".. so you can just ask the computer to remove a specific character encoding

"Remove all ASCII from this" I guess it would be.. can be done in excel/google sheets

or the terminal

[wanton raven]

https://tenor.com/view/good-day-sir-wonka-angry-gif-4938393

[mint heart]

lmao just wanna say it's funny several non-ASCII remover sites exist, but I can't find any that remove ASCII. maybe it exists on japanese google or chinese uhm version of google..

i think one of the harder issues is maybe the original file is an image file and not actually text you can copy/paste

i dunno if you can ask chatgpt to reliably change text in an image file into text you can copy and paste or if you should just use google's optical character recognition reader

cloud vision api I guess?

there are sites that do that too, found one and it seems accurate..

it's not my project but I'm just curious about it. Also I didn't know about workflows and it seems it costs money. I'm not sure how much 5000 steps is.

oh i thought you meant "google workflows"

is that really free in chatgpt without paying the monthly subscription?

uhm I don't see any way to give directories to chatgpt?

I can upload a file..

oh you mean for a file path.. I thought it was going to do all the work for me if I just told it to do something

I already had it write a python script awhile ago, which works.

i was thinking maybe there was something like a vscode extension i wasn't aware of

maybe co-pilot can do it tho

it uses either gpt or claude whichever i choose

[native swift]

Weird question, will ChatGPT run in the background on the app? Like let’s say I haven’t fully closed the app but I go to a different app will it continue to generate a response?

[low belfry]

yes

[icy oar]

Can deepseek analyse images

[brittle summit]

No

is deepseek really better than chatgpt?

I'm unable to use it since server is down

I'm mostly gonna use it for chemistry if anything

[uncut island]

https://x.com/reach_vb/status/1884142501671666035

Vaibhav (VB) Srivastav (@reach_vb) on X

NEW: DeepSeek Janus Pro 1B (Generate Images, Chat with PDF) running in your browser, 100% local, powered by WebGPU 🔥

Zero server costs, brought to you by transformers.js - try it out!

X · Jan 28, 2025, 7:32 AM

[boreal latch]

https://substack.tech-talk-cto.com/p/how-deepseek-disrupted-the-ai-giants

heres a high level overview created by featherless media team 3

[prime apex]

I was able to use it with images and it got everything right on my science tests which chat usually maybe gets a C on

[safe smelt]

This quote by Sam did not age well at all lmao

https://www.tomshardware.com/tech-industry/artificial-intelligence/sam-altman-said-startups-with-only-usd10-million-were-totally-hopeless-competing-with-openai-deepseeks-disruption-says-otherwise

Although I do see where he is coming from. This wasnt exactly a level playing field with Deepseek taking advantage of the work OpenAI already did. Actually beating OpenAI and coming up with your own new ideas would require much more.

[undone hollow]

Is deepseek working again orrr

[gray yoke]

i have chat gpt plus is this normal?

[safe smelt]

There’s a deepseek discord bot in a server from my profile bio. Seems fairly stable tbh

[brittle summit]

that link has expired

[fresh zealot]

why does the text sent in ChatGpt’s replies to me disappear once I refresh the page

[late magnet]

What channel has the featured posts from subreddit?

[green sparrow]

Lol that's interesting. Two in one. In the West it's always image gen as a separate thing

I did hear about Janus though. Wasn't sure fully about it

You guys talk more about DeepSeek in here than the real ChatGPT lmao

[spiral void]

how does deepseek r1 work, can i turn it off and will it be the same?

[stoic sleet]

What?

[spiral void]

it wont downgrade quality of writing i was asking

[pseudo peak]

what

[desert dagger]

no it turns to v3 which is lower

[solar trench]

Generated three random PhD cover letters with o1 Pro, Sonnet 3.5 and Deepseek R1 respectively
Which one would your guys prefer to use as template or reference

[boreal latch]

i hate all of these grifter 14 year olds who learnt the word ai 2 years ago

[tawny trellis]

https://openai.com/global-affairs/introducing-chatgpt-gov/

[dull tide]

Anyone know if the servers are ddown for chatgpt? Trying to work on something and i just keep getting the regenerate option

[brittle summit]

mines too it just says something went wrong

[wide thorn]

Currently just getting the message that something went wrong with generating a response too

[dull tide]

Ok so its probably down thats fine just was hoping I somehow didnt get cut off cuz im trying to make a Blackjack game for my discord bot lol

[brittle summit]

aight its all of us then

[toxic cedar]

I have the same

[daring scaffold]

2nd time when i needed it most lmao

[wide thorn]

When do you guys think that it'll be back this time?

[daring scaffold]

Last time it took 2 hours

[junior crater]

https://cdn.discordapp.com/attachments/1012371812002316381/1328524475263619113/IMG_8250.gif

[wide thorn]

I'm guessing it'll take 2 hours again or maybe 3

[dull tide]

Luckily I can work on other aspects of my bot while I wait. Or I can win like 1 fortnite match to hit Elite rank lol. Suffering or suffering are my options

[junior crater]

Me Every time the Chat Got servers are doen

[brittle summit]

hell na its 3am i cant just wait another two hours

[molten iris]

Side eye, why this server suddenly gets alive when something is wrong with CHATGPT?

[toxic cedar]

I hope not too long, I was in the middle of school work 😭

[reef hedge]

Did it go down?

[wide thorn]

Same, it is 9:08 pm for me, and I have work tomorrow

[wide thorn]

Yup

[reef hedge]

Ok dang, thanks

[glass nexus]

I think I broke the servers, abusing the AI with thousands of paragraphs. My bad y'all💀

[dull tide]

Someones cheating on their school work how dare they

[wide thorn]

Yeah, i think that you've murdered the AI with all your paragraphs lol

[brittle summit]

frfr i noticed that too 💀 i mean i joined this server only just in case whenever the server is down soooo.

[toxic cedar]

Noooo 🤣🤣🤣I would never dare

[glass nexus]

My bad, I just did a little trolling💀

[exotic harness]

GPT goes down:
Work stops ❌
Brings everyone to this server to ask if it went down ✅

[wide thorn]

I just use it for stories, so maybe my idk how many stories took part in murdering it too 😅

[molten iris]

At least it's not like c.ai's server that literally breaks my discord when it's down.

[supple wraith]

[toxic cedar]

"When it's down" = all the time

[wide thorn]

Tbf I would of just play WQAE until it is back, but that would mean leaving my cozy bed and starting my pc back up

So yeah, ima just chill in bed and wait for it to be hopefully back sooner than 2 or more hours.

[molten iris]

Explodes a place and runs.

[barren eagle]

Hey there, is it just me or is chatgpt completely down? can't even function

[toxic cedar]

Yep

[brittle summit]

thats why everyones here right now

[wide thorn]

So....did anyone bring the snacks and drinks for, god knows how long, it'll take to come back to life?

[dull tide]

thats the only reason people are talking. Genius marketing strategy brings the folks to them

[barren eagle]

great, just when i needed it for something extremely urgent 😮‍💨

[dull tide]

I didnt. I was programming Blacksjack into my Yugioh bot cuz why not

[ruby venture]

Hey, is chatgpt down rn?

[brittle summit]

use deepseek i guess ☺️

[molten iris]

My wife is so hot fr.

[junior crater]

Your not the only who makes up stories my brotha

[amber gorge]

normally how long does it take to restore?

[dull tide]

3.5 years

[wide thorn]

Nice

[amber gorge]

lol kms

[molten iris]

No, it's 10 billion years.

[wide thorn]

At times it feels like 10 years tbf depending on how long it is down

It is back on my side!

[brittle summit]

oh shi really ??

[supple wraith]

its back ! we can leave now

[wide thorn]

Though now I gotta wait for the o4 to be back in 20 or 19 mins lol

[brittle summit]

aight bye yall

[dull tide]

See yall next crash!

[wide thorn]

Bye, until the next outage!

[toxic cedar]

Bye 👋🏼

[junior crater]

See y'all next crash Too also

[dull tide]

Damnnnnn the crash deleted the last chat I was working with and I was decently in the middle of working on a new command

[brittle summit]

o3 launch today is true?????

[ruby venture]

Is there a way to recover the chat?

[dull tide]

not that I know of. Ill just have to start it from scratch im just trying to program Blackjack into my bot lol

[stoic sleet]

no

it's probably tomorrow or maybe next Tuesday? Only for o3 mini

[brittle summit]

Is o3 mini better than o1??

[stoic sleet]

In coding and math. The rest remains to be seen

It’s much much cheaper

[brittle summit]

Damn if cheaper than it's going to be game changing 💯🔥🔥🔥🔥

[stoic sleet]

o3 mini medium is cheaper than o1 mini and overperforms o1 by about 100 elo on codeforces (bigger than it looks)

$20/month will come with 100 messages a day

[tender radish]

this may be a dumb question but

how can i get AI to address word comments?

[brittle summit]

Best time to be alive 💯🔥🔥

[tender radish]

For example I might get a report back marked up like this

But when I put it into deepseek it says it can't read word comments. Is there a way I can ask it to do this correctly?

[mint heart]

lol i saw the super long thought thing on the sub about asking deepseek the probability of answering a question. I got it to do similar by asking it to solve a Millennium Prize Problem.

only gave me 161 seconds of thoughts though, so almost 3 minutes of thinking?

reddit post only showed 120 seconds of thinking before it cut off

maybe if i asked it to solve p=np it'll go on for longer

[stoic sleet]

Deepseek can't handle files at all

you'd have to have the comments in the text. ChatGPT can convert it into something Deepseek can read

[mint heart]

p versus np only got it to think for 153 seconds

[safe smelt]

I don't quite understand that "upcoming" cost. Makes no sense to me tbh. Better newer gpus also mean more compute for same price, but this estimation seems out of thin air wanting to build the hype

models are getting smaller, not bigger. And we already using the entire internet of data pretty much. RL for reasoning models is a relatively new concept, but I don't see how that would counter models getting smaller or it even being comparable to the amount of data used in pretraining.

[tender radish]

So I can upload the doucment into ChatGPT and ask what?

[steep pier]

chatgpt try to not get downtime challenge
99.999999999% fail

[brittle summit]

Where is o3 pro

[low belfry]

end of the month is the friday

they said before the end of the month

they should release it at 5pm on friday

[stoic sleet]

Idk

I’m asking too

[proper sorrel]

Are the free bots here capable of data analysis?

[wanton raven]

what kind of data analysis?

[velvet turtle]

For the atom people who think China is watching them

[wanton raven]

Trentbot for example is always watching and recording and analyzing and j̵̙̅u̴̍͠d̴̞͆g̶̒̑i̸̊̌n̵̔̀g̶̭̓

[dull tide]

So I just hit my Plus limit on working on a singular command because every small change I try to make the AI just removes a massive chunk of the file.

[low belfry]

have it work on just a small section at a time

[proper sorrel]

Math type shit. Uhh statistics

[vivid drift]

https://bsky.app/profile/syrussolojin.bsky.social/post/3lgvmvevhpc2r

Syrus Solo Jin (@syrussolojin.bsky.social)

"Generative AI is experimental," the disclaimer notice indicated in small text, as OpenAI earns a contract to run Medicaid

Bluesky · Jan 29, 2025, 6:45 PM

[velvet turtle]

a Chinese twitter?

[copper storm]

So this aged like milk.

[vivid drift]

Nope

USA Twitter

[velvet turtle]

What is the tweet you sent about? I don't want to login to see it.

[wanton raven]

imagine not knowing about bluesky. I don't get the reasoning behind "sign-in required" posts though

[stoic sleet]

with o1 or 4o?

[dull tide]

4o

[echo blade]

where here would i post "questionable" chat gpt answers

[acoustic glen]

hey is anyone’s chat GPT responds feel more personal in a way? like the way it use to respond before versus the chats i had today are like night and day. before it was like i was definitely talking a AI now it’s like i am still talking to a AI but it seems more like i am speaking to a actually person over text? if i can send a screenshot of a part of our convo i will, just lemme know if i can! ❤️

[echo blade]

cuz i just told chat gpt to only talk in emojis and slang, and it decided to call hitler a sigma male

im just gonna put this here LOL

[acoustic glen]

ur iconic ty ty ❤️

[echo blade]

brainrot chat gpt is now my biggest source of entertainment

[vast sequoia]

Is deepseek r1 that is available on the website the full r1 that the benchmarks say rival o1? Can’t really find info online, it’s really unclear. Is it distilled or a watered down version?

[dreamy ivy]

OpenAI crying publicly about deepseek using their models responses to train is wild to me

[stoic sleet]

Yes chat.deepseek.com when you enable deep think should be around o1 preview

[green sparrow]

Yeah. I think it's a mix of people talking more casually to AI chat bots over time as they get more comfortable with them, and AIs reciprocating the same style back

[acoustic glen]

word word. very cool i must say. makes the convo more engaging, for me at least lol

[mint heart]

tbh I trust china deepseek more than I trust open ai <<, if I ask a bad question out of curiosity I might have the fbi knock on my door. But if I ask the chinese government a bad question, worst thing they can do is blackmail me to become a spy, but i'd be worthless as a spy, so they wont.

[wanton raven]

just gonna hang on to this

[deep jackal]

AGI mode

activated

[green sparrow]

I just noticed that button on my Microsoft Store app version, which is strange because they seem to always be behind the Mac version

They must have been able to update that more intelligence button without releasing a new app update

[green sparrow]

Nah that's probably just the reasoning model o1 harhar

Guys if DeepSeek gets proven 100% as stolen (distillation), it probably is going to get removed from the App Store/Google Play Store

[mint cosmos]

Even better, if you self host it, which you can do with deepseek because if it's open source, the worst that can happen is you getting misinformation.

[mint cosmos]

stolen from where?

[green sparrow]

From the models from OpenAI

[mint heart]

isn't the online model like 600 B permamerters or something? so I'd need at least 600 GB of vram to get a model of equal usefulness?

[green sparrow]

Look up the latest news regarding open AI alleging that deep seek stole their data

[mint cosmos]

oh okay

well if it is, then it's still a good thing

because the licensing difference

and optimization

[mint heart]

lol, this kinda seems absurd and conspiracy-theory-ish... but deepseek might train on discord data. ..

[green sparrow]

Bro discord training data would be awful lmao

[mint cosmos]

I don't think so

[mint heart]

someone came up with some ML training program, about 5 months ago and I asked it to provide examples of use cases..

I joined the server and one of the use cases is exactly the same use case someone said "id like to try and use this program for"

could just be coincidence

[green sparrow]

Discord training data: filled with misinformation, typos, and slang

[mint cosmos]

Theories that I believe:

• Deepseek caused the few hours of ChatGPT and claude outage
• Deepseek has a backdoor in it's weights, even if they're open, you can't understand a neutral network
• It's trained on OpenAI's models, which is why it was supposedly so cheap to make

[mint heart]

all LLMs train on random online data

so typos don't affect it

[green sparrow]

I just don't think they would do that

[mint heart]

just weird concidence i guess

[green sparrow]

Yeah

[mint heart]

Also, I think discord data would be useful to train an LLM

just as useful as reddit, if not more so.

So I wouldn't be surprised if some company contacted discord for chat access on all their servers to train a LLM

lol another coincidence is a "Fantasy Football Draft Advisor" <.<

[worthy roost]

comparing deepseek and chatgpt lol

[cedar wing]

Hi Chatgpters! (the hell idk what is this word lol)

[stoic sleet]

Someone tried it here

It might make sense on some highly technical and moderated discords but it’s just not worth it

Reddit is a much better format. Presumably Reddit AI training considers upvotes too

[steel hedge]

Sorry if this is too recurrent but i thought it gave me some explanation about it's restrictive guidelines

[stoic sleet]

Interesting 🤔

[odd epoch]

Hey peeps, I was wondering other than Copy AI, Claude Ai and Chatgpt, are there any alternative good platforms (free) that can help with the drafting of my project proposal for my company?

[copper storm]

Your government can still potentially monitor what you say to the Chinese AI, its only truly private if you run it locally without an internet connection, which you can since it's open source.

[wheat turret]

Hey guys, long time lurker here. Today I got a message saying: "You have 25 responses from o1 remaining.". I have a paid subscription and never saw this message before. Is it new?

[copper storm]

No you have 50 weekly o1 messages or so on the plus plan.

[wheat turret]

Gasp

[brittle summit]

USA is a falling empire. CHINA won the artificial intelligent race and everyone is seeing that USA is a cry baby with double standrds

[exotic kiln]

I'm honestly tired of the 4o mini can't do attachment BS, is there an AI chatbot that can do attachment 24/7 no limits

Why doesn't it work like that

Or maybe even something that can run locally

[lapis quartz]

Get an API token and you can download one of many open source UI suites and also set up your own RAG database locally. Then you can attach files until you get sick of it

I use LibreChat with a RAG api by the same dev, based on pgvector postgres embeddings db

[brittle summit]

rekt

[mint cosmos]

I thought it wasn't possible to make me antagonize deepseek and side with OpenAI until I saw your message

OpenAI is terrible, everything is closed source, there for profit, even if it's indirect they are literally selling everyone's data.

Deepseek is open source, which is good but they probably didn't do all the work

they used OpenAI's model to train theirs, which doesn't mean that they are winning at AI, it just means we have a slightly worse open source clone of o1 now

Which is a huge w for everyone except OpenAI

Also for Americans, Everyone includes Americans

[celest cypress]

u got rickrolled

[radiant zephyr]

Live

[celest cypress]

open AI was only claiming it, which is cannot be entirely trusted

[quasi egret]

Guys how do I view the very start of a topic in chatgbt? And view it all? I've gone to the very top and it only went to a recent question in the topic?

[smoky sorrel]

why does some citation urls have the utm_source=chatgpt.com and some don't?

See these two screenshots for example (on the same answer)

[last axle]

is deepseek better than chatgpt

[vivid drift]

In my testing no, but in benchmarks yes

[brittle summit]

no.

[stoic sleet]

better than 4o? Probably. Better than o1? No

[south prism]

when is o3 mini coming out

[stoic sleet]

when I decide to release it

[mint cosmos]

yeah but logically think about it, and test the similarities. it's undeniable that they at least used partly OpenAI's models as training data

[mint cosmos]

Definitely not

In my testing it's much worse for every day use

it's probably just because I got acclimated to ChatGPTs style

[south prism]

ok gotchu can u realease o3 pro for me

[stoic sleet]

no want to give China a chance to catch up first

maybe can do it just for you?

[brittle summit]

When is o3 coming out

[stoic sleet]

Feb 23rd

March 2nd for o3 pro

[brittle summit]

https://9to5google.com/2025/01/30/google-ask-for-me-lab/

Bruh imagine an ai calling you to ask about prices and stuff.

Tbh my pixel assistant for screen calling sounds near human

[stoic sleet]

apple needs an answer to this

[brittle summit]

Bruh it didn't record the audio

There

[brittle summit]

@stoic sleet listen to that.

It does that before even reaching my phone

[small birch]

Is anyone else having a problem where the custom instructions when you make changes and click save it doesn’t save your changes?

[green sparrow]

Stolen data is fun and makes you look powerful
Until you get exposed

[clear atlas]

Hey !

Someone know if we can have chatgpt with carplay ?

With shorcut iphone ? Or something

[dark eagle]

can someone help me with coding a robot but getting the whole thing from chatgpt ( i will be using python on an android) to host robot

[brittle summit]

@dark void are you adding bots to the discord

These people feel very bot like

[dark eagle]

https://tenor.com/view/uhm-uhmm-yes-huh-uhmmm-gif-23719974

[dark void]

everyone i dont like is an evil foreign robot

[pure aurora]

where is o3

im ready

[stoic sleet]

probably coming on the 4th

for o3 mini. o3 not sure

[pure aurora]

yesterday they said tomorrow, they lied 😦

[brittle summit]

Where is o3 mini

[celest cypress]

can be

but probably not much

[stoic sleet]

it will come on the 4th probably

[pure aurora]

not yet published

[stoic sleet]

they didn't say it would be today

[pure aurora]

how you know ?

[stoic sleet]

tuesday or thursday launch is most likely

I don't see it taking that much longer but we will see

[pure aurora]

im too excited for agi

[celest cypress]

but like before, deepseek was sometimes just claiming it

the llm

[safe smelt]

yeah but they did say it was coming at the end of January. Tomorrow is the very last day of January 💀

[stoic sleet]

It’s not coming in January

[safe smelt]

Well then so much for all this talk of Deepseek supposedly forcing them to speed things up. They can't even meet their old deadlines

[vivid drift]

It was trained on outputs of chatgpt using the api. Nothing was stolen. Microsoft found the api key they used and disabled it

Outputs of Ai can't be copyrighted

[dull tide]

Is it just me or is the AI on Chatgpt getting dumber lol. Before I had new issues getting a file update through it. But now it always removes something random every time I try to work on an update

[stoic sleet]

o3 releasing tomorrow 👀

[quasi egret]

Guys how do I view the very start of a topic in chatgbt? And view it all? I've gone to the very top and it only went to a recent question in the topic?

[humble magnet]

not dumber. just more lazy bc its being directed that way by openai to reduce costs (cus corporate greed :3 )

[solar egret]

Can PLEASE someone in here that has ChatGPT Plus do me a little favour? I want to know if it works out. It's just an Excel File that I want to modify for my personal needs but I keep getting the message that I reached the data analyse limit and need to purchase ChatGPT Plus. It's literally an Excel File I wrote the prompt in German where I explain everything I want to modify and I'm wondering if chatgpt can actually do that task

I would appreciate it soooo much if someone can message me and do me that little favour 🙏

Whenever I create a new account it does everything I want and mid way it says „limit reached for data analyse please purchase ChatGPT Plus“

[brittle summit]

I'm really tired checking the twitter everyday and people saying tomorrow tomorrow tomorrow, this is 31st for me, i HOPE it releases today. 😭

[brittle summit]

o3 releasing tomorrow buttt

You should buy trentbot coin trust

[mint heart]

lols

[hard shore]

Anyone know a good ai for graphing or like Pictures of graphs to answer?

[mint heart]

i can only think chatgpt maybe pro? and only if you instruct it to write in code

[hard shore]

Chat gpt 4o wasn’t working to good for my domain and range graphing

[mint heart]

don't think any do natural language to direct math

[hard shore]

mmm

[mint heart]

but you can do natural language to code, to math

like python or something

[hard shore]

I’ll try that

[mint heart]

oh and several websites popup when i typed in "graph ai" into google. Not sure if they're good or will work how you want..

was thinking an ai wrapper service might exist for such a thing, and I guess it might. Maybe?

[hybrid vapor]

GPT working for anyone else?

[molten iris]

No.

[sinful nebula]

is chatgpt broken for anyone else?

[quartz spruce]

yep

[sinful nebula]

I was getting help with my game 😭 hopefully it won't be down for long

[desert dagger]

it's DEAD

CHATGPT IS DEAD

[quartz spruce]

stop

[desert dagger]

https://kimi.ai/ is better than chatgpt and works while chatgpt is completely dead

https://kimi.ai/chat

[harsh crown]

iis chatgpt broken for you guys too i just joined to ask

[quartz spruce]

yep

[harsh crown]

ok good to know its normal unleast

[quartz spruce]

someone in the openai community said projects were back to not reading project files, I went to check, and chatgpt wasn't working right.

[quartz spruce]

are there no human mods in this server?

[desert dagger]

only animal mods

[lone lily]

Seems to be fine now

[quartz spruce]

no o3 for me

what's that have to do with anything?

[desert dagger]

nothing

[quartz spruce]

I said I don't have the o3 model yet

[sinful nebula]

what about images?

no I meant giving it an image

ah ok! so seems like it works now

[sinful nebula]

Have ChatGPT gotten dumber recently?

more spesifically 4o

[quartz spruce]

Any idea when this is dropping? Or was this a great big hoax?

Are you trolling?

you said you made it up in your head!

Because as a writer, I would benefit more from chatgpt being able to recall details from past conversations.

[peak kraken]

https://cdn.discordapp.com/attachments/1283127748763713556/1315082111857983500/togif-1.gif

[green sparrow]

Guys I think they got rid of that "use more intelligence" button that randomly popped up at around midnight yesterday

Is anyone else frustrated and confused why you can't share your desktop screen with the ChatGPT app from Microsoft Store? It doesn't make any sense

[teal wren]

o3 or o3 mini today, and is it only pro plan?

[wanton raven]

openai claims it loses money to chatgpt pro, imagine how much more it would lose if suddenly all conversations have "unlimited" context

[stoic sleet]

o3 mini will be coming to the free tier

[humble magnet]

openai is a pro-israel pro-trump garbo dump

excuse my english

[wanton raven]

that hardly seems relevant

[exotic rune]

Has anyone tried it before?

[humble magnet]

it is relevant bc we expect chatgpt to give objective and accurate information?

[wanton raven]

none of us expect it to do that unless you're brand new to it

[humble magnet]

Im not new and i still expect that

[wanton raven]

Yet you haven't seen the "ChatGPT can make mistakes" on the bottom of the page?

[sinful nebula]

tfw gpt 4o is kinda stupid, o1 is limited and deepseek has busy servers

[stoic sleet]

AI Studio

[sinful nebula]

Is it free? And good for coding

[stoic sleet]

Yes

Gemini Flash 2 Thinking is amazing and free/virtually unlimited

65k output limit it’s insane. Better than Deepseek for sure

[brittle summit]

It's literally the best thing given for free in this entire world for sure.
There are some old lectures in youtube that the professor either forgot to upload the notes of or the notes url got corrupted.
I literally downloaded the whole youtube lecture, pasted it into AI studios and told it to code me latex that is same as the lecture, and with a bit of tweaking AND IT WORKS, SAVES A TON OF TIME NGL. I was literally screenshotting every page while taking the lecture before and teacher not moving from the screen pisses me off. 😭
The output is very identical and I can just open up goodnotes in my ipad and watch the lecture while making marups live during the lecture. 😭

hehe

I'm just in love with this thing

[stoic sleet]

It’s just the best

[brittle summit]

literally, if they really ship ai studios as an app, it WILL be the number one on the app store for sure. 😭

[analog token]

very cool - could you post the lecture youtube link?

[exotic kiln]

This DeepSeek fearmongering is still ongoing, with a new meme everyday, shoot me dead if this is not a psyop

The Chinese are very good at stealing tech, you think DeepSeek will be what it is today without OpenAI?

Who even believes some hogwash you can build an AI with sub 10 million dollars?

[brittle summit]

Sure.

https://youtube.com/playlist?list=PL_A4M5IAkMafmea0RnicbuLD6Ly2EIr0X&si=p-wxcRtyBBTXXQU-

It is in Hindi though.

I have extracted notes from lecture number 2, 3, 4 so far but I have to tweak them a little bit, gonna do it later.

Here is an example from one of the lecture. Gotta fix the latex later. And yes, it understands what the teacher is saying even though it is in seperate language.

(I cleared out the prompts and outputs generated by AI studio by clicking on flush button lol, otherwise I would've provided that too.)

[lyric comet]

Why does ChatGPT work so terribly? I can't even upload a .txt file. I uploaded a few .txt files, and now it says the file is broken, can't read it, and asks me to upload again and again. I paid for the Plus plan.

---

I work for about two hours in a new chat (New Chat), and then it stops working properly. I have to start everything from scratch in a new chat, and ChatGPT doesn't remember anything from before. So I have to spend hours explaining the project again, and then it stops working again—neither the files nor ChatGPT's reasoning function properly. I paid for the plan, so what could be the issue?

[brittle summit]

I don't know if it supports .txt format, it is saying "File is empty" as an error pop up for me.

[brittle summit]

So it does support .txt format, hmm?

[lyric comet]

In a new chat, I can upload .txt files again, and everything works. But after some time, I can no longer upload files, and nothing works.

[brittle summit]

Have you reached your 4o model limit or something? You can send like 80 messages to the GPT-4o model in 3 hours if I'm correct.

[analog token]

@brittle summit thanks :)

[brittle summit]

No problem!

[brittle summit]

What I think is that you actually reached your 4o limit and the model automatically switched to 4o-mini which may not support .txt format or something?

[lyric comet]

the limit reseted

no no i'm not that stu*pid 😄

The next day, nothing works—I can't upload data, and GPT doesn't remember what I did. I have to start over in a new chat again.

[brittle summit]

I can feel you, sometimes my ChatGPT just doesn't support files during projects I'm working on which is very frustrating. It just doesn't let me upload files and the 4o model gets really slow.
Sometimes a specific file ceases to upload while others can very perfectly.

[brittle summit]

Try uploading some photos or pdf documents to check if uploading external files is not working in general?

[lyric comet]

image works

[heady crystal]

bro my stuff is broken i cant se my previous chats with chat gpt

[spiral void]

these chinese ai will definitely get tiktok treatment in the us

[safe smelt]

you realize that China is the only one with the blocking thus far, right?

Chinese government explicitly and without any negotiations banned chatgpt website for no good reason lol.

[vivid drift]

Probably not

[spiral void]

cuz its american and the americans will def steal data from chinese and vice versa with the chinese ai lol

[wanton raven]

It's because it provides access to forbidden information

[safe smelt]

It's not really a vice versa. Only 1 country is actively blocking everything

I'm not really a fan of tiktok at all, but I'll admit even blocking just that alone could be opening pandora's box. Under true democracy nothing should be blocked

it's not really government's business what I decide to do with my data or what I choose to access, fundamentally. Though this comes with a caveat that fighting for user data privacy by negotiating with the companies is very important still

[copper storm]

Damn.

I think everyone in china uses a VPN anyway though.

[south prism]

this isnt relevant 😄

[mint heart]

deepseek down?

[brittle summit]

It is not working for me too i just checked.

Like kinda works but is really really slow.

[mint heart]

kk thx, i uploaded a pdf and that's when it stopped working, wasn't sure if the pdf was too big or a coincidence

sucks I need at least 120 GB of ram to use it locally, or so someone said on the localllama sub

[pure aurora]

in May nvidia will release a personal 3000$ pc with 128GB of unified memory. Kinda like vram but 2-3x slower

With it you can run 200B models probably, with 2 of them you can run llama 405B

[mint heart]

Im not gunna buy another $3000 computer for awhile.

like 10 years

[stoic sleet]

It’s starting at $3000 for up to 128gb

[pure aurora]

oh up to ...

[stoic sleet]

Yeah I’m guessing $3500 for the 128gb? Also there’s concerns the memory really won’t be that fast to the point it could really hamper models

[pure aurora]

are you pro though ?

proof or it didnt happen

[stoic sleet]

You need to verify to get the role

I had to send my receipt to get the pro role

[mint heart]

deepseek still seems to be down? maybe openai is ddosing deepseek so people can be interested in their new model releases

if chatgpt turned on a thinking option for free users and it looked like deepseek's i'd stick with chatgpt 100% of the time <.<

[analog token]

yeah its down for me

I don't recall ever seeing a website this popular (ie #1 app for several days now) have this much server trouble

[mint heart]

it started to crash as of 2 hours and 30 mins ago for me

which is the same timing as chatgpt's new models came out I believe

[pure aurora]

US ddosing china 😏

[gritty cipher]

AI won't replace music production 100%. Song is AI, but I recreated every single instrument, only the vocals are original.

[green sparrow]

Chats got leaked from deepseek

[gritty cipher]

There are some people who are really against AI for music, or generative AI in general making lazy people who put zero effort, but in my case It took me a couple days the make the song, And I had to listen to find out the chords, the notes,etc

[slender knoll]

Ai just helps in music right it doesn't create one

[steel hedge]

Got him.

[smoky arch]

Hi

[brittle summit]

Is the role out for o3-mini started yet?

[vital idol]

supposed to be 10am PST which has already passed

[runic vale]

Where’s o3 mini?

[stoic sleet]

maybe o3 mini was the friends we made along the way?