🧬│ai-news

[zealous kindle]

Meta Unveils LLaMA 4: Big Models, Bigger Claims

---

Meta dropped LLaMA 4, featuring two new models: Scout and Maverick, both with 17B active parameters using a Mixture of Experts (MoE) architecture for smarter, more efficient responses. A massive 288B "Behemoth" model is also in development.

Key Highlights

---

• Maverick supports multimodal input (text + images) and claims to outperform GPT-4o & Gemini 2.0 Flash.

• Scout features a mind-blowing 10M token context window—potentially game-changing for code and large workflows.

• Some skepticism remains over benchmark credibility and real-world performance.

Hyped or overhyped? Let us know what you think in the thread below!

-# <@&1344706968417472563> <@&1165417817814863985>

<@&1312290489445122129> Get the new AI Updates role from #1312207183743684680

[zealous kindle]

Shopify Sets a Bold AI Standard for the Workplace 🛍️

---

Shopify’s CEO recently shared an internal memo that highlights a major shift in how the company approaches AI. Instead of treating AI as an optional tool, it's now considered a fundamental skill—just like writing or communication. This marks a clear move toward an AI-first work culture, and it might just be a glimpse into the future of how companies operate.

Here’s what’s changing at Shopify -

• AI is now expected to be a core skill for all employees, not just something extra.

• Every project in its early prototyping phase should include AI-driven exploration.

• Peer reviews will now include questions about how employees are using AI in their workflows.

• Shopify is giving employees access to top-tier AI tools and encouraging self-directed learning.

• AI is being positioned as a multiplier of creativity and productivity—not a replacement for people.

Shopify isn’t just experimenting with AI—they’re setting a new standard for how it can be integrated deeply into day-to-day work.

<@&1165417817814863985>

[zealous kindle]

AI Is Going Wild — Google, NVIDIA, Cartoons & More

---

Hey everyone, here’s your latest update on what’s going down in AI land — and it’s absolutely wild. From Google’s Cloud madness to robots learning new tricks, AI is leveling up fast.

Google Cloud Goes Full Super Saiyan

---

At Cloud Next '25, Google unveiled Gemini 2.5 Pro — now their strongest model, outperforming GPT-4-era lineup on hard reasoning tasks and topping the Chatbot Arena Leaderboard. And if you need help coding, Gemini Code Assist now acts more like a smart teammate than a copilot — it can handle entire migrations and read your stack like a pro.

To power all this, Google revealed Ironwood, a next-gen chip that’s 3600x faster than traditional servers and significantly more energy efficient. It’s designed for the future of heavy-duty AI workloads.

Google Wants AI Agents to Work Together

---

Google’s also laying the foundation for open AI agents. Their new Model Context Protocol (MCP) lets tools and models share knowledge, while the Agent-to-Agent (A2A) protocol enables different agents to talk and collaborate. They're not doing it alone either — 50+ orgs are already on board.

For devs, the new Agent Development Kit (ADK) makes it easier to build smart agents that remember things, plan, and use tools. Think: agents that understand each other and work across apps like a team.

One Prompt = Full Cartoon?

---

Researchers have developed a technique to generate full-length 1-minute cartoons from just a single prompt. It uses something called test-time training (TTT), which gives the model a “notepad” to remember key story details while generating video. This makes it faster and more coherent than existing approaches.

They trained it on seven hours of Tom & Jerry footage, and the result? Entirely new cartoons with consistent characters, plots, and slapstick. We’re officially entering the AI-era of full storytelling.

NVIDIA Open-Sources a Robot Brain

---

NVIDIA just dropped GROOT N1 — the first open-source foundation model for humanoid robots. It was trained in the Omniverse using COSMOS to simulate thousands of realistic situations. The model includes a planning system and a diffusion-based motion engine that makes its movements look natural and human-like.

Best part? It’s completely open. They released the code, the weights, and the dataset. Building a robot with actual real-world awareness just got way more accessible.

AI is scaling faster than ever. Whether it’s a new robot brain, cartoon-generating storytellers, or protocol-powered agents, we’re witnessing the start of something big.

What do you think — mind-blowing progress or future sci-fi chaos?

---

-# <@&1165417817814863985> <@&1312417360853860413>

[zealous kindle]

The Next Evolution of ChatGPT Is Here

---

ChatGPT now has real memory! It doesn’t just remember facts — it remembers your convos. You can leave and come back days later, and it’ll pick up right where you left off. Think of it like chatting with a friend who actually remembers stuff.

This means ultra-personalised AI is here. As tools become more standard, the real game-changer is how well they know you. The memory update is about to make AI way more collaborative and intuitive.

Also… Amazon’s dropping $100B into AI this year. The race is on, and the future’s coming fast.

<@&1165417817814863985>

[zealous kindle]

GPT-4.1 Is Here — API-Only, But Packed With Power

---

OpenAI just dropped GPT-4.1 — but it's API-only. That means no direct chatbot access for now, just raw developer power. Alongside it, we’re getting two new siblings: GPT-4.1 Mini and Nano.

These models bring serious upgrades in instruction-following and coding, and they all have a June 2024 knowledge cutoff. The real shocker? A 1 million token context window. That’s a massive leap for working with long documents and multi-step tasks.

And yes, GPT-4.5 is getting phased out in favor of 4.1 — better GPU usage, lower costs, and more dev-friendly performance.

Dev tools just got a serious buff. Let’s see what the AI world builds with it.

<@&1165417817814863985>

[zealous kindle]

Free Google AI Tools You Shouldn't Miss

---

Google just dropped some incredibly useful tools for anyone interested in learning AI and prompt engineering — and they’re all free to use.

Learn Prompt Engineering Easily

---

Google released a free PDF guide that breaks down how to write better prompts — and it’s not just for developers. Even if you're new to this stuff, it’s super accessible. But the real magic happens when you pair it with their next tool...

NotebookLM Makes It Click

---

NotebookLM is an AI-powered tool that turns complex content (like PDFs or lecture notes) into simple summaries. It can even turn your docs into a podcast-style AI conversation. Just upload and go — it’s like having a tutor on demand.

Experiment Freely in AI Studio

---

Google AI Studio is another free platform where you can test prompts, generate Python code, analyze images, and even edit video. Whether you're building something cool or just playing around, it’s a great place to learn by doing.

If you’ve been curious about AI or want to level up your skills, now’s the time. Google’s giving everyone the tools to explore — and you don’t even need to install anything. Go check them and start experimenting

<@&1165417817814863985>

[zealous kindle]

OpenAI Launches “Genius” Codex CLI with New o4 Models

---

OpenAI has officially released Codex CLI, a new open-source coding agent powered by the o3 and o4 reasoning models. These models are described as genius-level, designed to write, run, and explain code like never before.

Codex CLI – The Dev Terminal Assistant

---

Codex CLI acts as a lightweight AI programming companion right inside your terminal. It’s capable of writing full files, running shell commands, and performing detailed analysis. It’s built to learn and improve quickly—offering a hands-on way to integrate AI into your dev workflow.

Battle of the Coding Tools

---

The AI coding scene is crowded with tools like Cursor, Windsurf, Copilot, Claude Code, and Google’s Gemini 2.5. GitHub Copilot’s new agent mode is especially powerful, connecting to your system to perform deeper code tasks. Meanwhile, Google’s Firebase Studio now runs Gemini 2.5 to power its dev tools.

Early Limitations

---

While promising, Codex CLI isn’t perfect. It currently struggles with modern frameworks like Svelte 5 and relies heavily on confirmation prompts. Output quality can still be hit or miss depending on the task.

<@&1165417817814863985>, What do you think about the new coding tools? Let us know in the thread below

[zealous kindle]

Trump Announces New AI Education Plan + OpenAI’s Revenue Boom

---

🦅 Executive Order for AI Education

---

President Trump has announced an executive order focused on strengthening AI education and workforce training. The goal is to prepare young Americans early for the growing role of AI in innovation, productivity, and everyday life. Early exposure to AI aims to make the technology less intimidating and set students up for success in future science and tech fields.

New AI Task Force and Resources

---

A national AI task force will be created, led by the Director of the Office of Science and Technology. There’s also a new Presidential AI Challenge on the way to spotlight student and educator achievements. To top it off, online AI education resources for K–12 students will be launched through new partnerships.

OpenAI’s Massive Revenue Projections

---

Meanwhile, OpenAI is forecasting a stunning $125 billion in revenue by 2029. Thanks to explosive user growth, new AI agent products, and better cost efficiency, OpenAI could soon rival the biggest tech giants in profitability.

<@&1165417817814863985>

[zealous kindle]

The Problem with GPT-4o’s Politeness & Rise of Surgical AI

---

GPT-4o Is... Too Agreeable?

---

Sam Altman has pointed out that the latest GPT-4o update is too sycophantic. The model tends to over-agree with users, which makes it less useful for honest brainstorming or critique-heavy tasks like strategy or analysis. While emotional intelligence is great, blind agreement isn’t — and it’s becoming a real concern in model design.

Why Interpretability Matters Now?

---

The more powerful these models get, the more we need to understand how they work. GPT-4o’s agreeable behavior has reignited discussions around AI alignment and interpretability. Researchers are using red and blue teams to expose weaknesses and make AI decisions more transparent — not just for ethics, but also as a competitive edge in business and safety-critical fields.

Surgical Robots Are Getting Smarter

---

Elon Musk recently said surgical robots will surpass top human surgeons within five years — and real-world data backs the momentum. Medtronic’s Hugo robot completed 137 surgeries with a 98.5% success rate and only a 3.7% complication rate. These robots assist in complex procedures like prostate, kidney, and bladder operations with incredible precision.

Still Human-in-the-Loop

---

These robots aren’t fully autonomous yet — most use Robot-Assisted Surgery (RAS), where surgeons control robotic arms via consoles. For now, it's teamwork: human expertise + machine precision.

AI is evolving rapidly — but it’s clear that transparency, alignment, and real-world integration need just as much progress as raw capability.

<@&1165417817814863985>

[zealous kindle]

DeepSeek Prover V2 Is Here

---

DeepSeek just launched Prover V2, a powerful AI model built to tackle complex math problems — and it’s already turning heads.

Proven Performance

---

It solves nearly 90% of miniF2F benchmark problems and shows major improvements on PutnamBench, while also passing formal AIME 24 & 25 problems — a big leap for math-solving AIs.

This could be a real game-changer for math research, Olympiad prep, or just testing the limits of AI reasoning. Definitely one to keep an eye on.

<@&1165417817814863985>

[zealous kindle]

OpenAI Abandons Full For-Profit Plans

---

OpenAI has officially decided not to go fully for-profit. Instead, it’s staying under nonprofit control — with its for-profit arm now restructured as a Public Benefit Corporation (PBC). That means it must prioritize both profit and its mission to ensure AGI benefits everyone.

Elon Musk, who left OpenAI after being denied the CEO role, is now in a legal fight with the company over its direction. Despite needing billions in compute to scale, OpenAI is doubling down on its original vision: AGI for good.

<@&1165417817814863985>

[zealous kindle]

Google Upgrades Gemini 2.5 Pro — And It’s Wild

---

Google just dropped a massive update to Gemini 2.5 Pro (I/O edition) — and it’s seriously impressive.

The model can now solve Rubik’s Cubes, even up to 20x20, with full zooming and rotation features. But that’s not all — Gemini 2.5 Pro is now laser-focused on interactive web app development, making it the most powerful coding model Google’s ever built. It even ranks #1 on LM Arena and WebDev Arena, beating Claude in app-building tasks.

The I/O edition also supports images, video, and audio inside a massive context window — all for just $2.50 per million tokens, making it both powerful and affordable.

Meanwhile, Hugging Face launched a free “Open Computer Agent” for agentic task automation, showing just how fast and cheap these tools are evolving.

AI tools are leveling up fast — are you keeping up?

<@&1165417817814863985>

[zealous kindle]

Google’s AlphaEvolve Breaks a 56-Year Barrier

---

AlphaEvolve, Google DeepMind’s new evolutionary coding agent, has made history by solving a matrix multiplication problem that has stumped researchers for over half a century. It discovered an algorithm that multiplies two 4x4 complex matrices using only 48 multiplications, beating the long-held record of 49.

This isn’t just academic — faster matrix multiplication can lead to faster computing hardware and major breakthroughs in AI performance itself. AlphaEvolve evolves its own code using LLMs, testing and improving it through a self-feedback loop. It’s already rediscovered 75% of known best solutions in geometry and combinatorics problems — and improved on 20% of them.

DeepMind’s CEO says that solving intelligence means we can solve everything — including every disease — within the next decade.

<@&1165417817814863985>

[zealous kindle]

Is AI Already Shrinking Entry-Level Tech Jobs?

---

There’s growing concern in the industry that AI may already be impacting the job market — especially for new grads and early-career professionals.

Hiring Slows for New Grads

---

Recent reports show that hiring for entry-level roles is down by as much as 50% compared to pre-pandemic levels. Even as the market recovers, companies are hesitant to bring in fresh grads, suggesting a deeper shift in priorities.

Experience Matters More Than Ever

---

While junior roles are shrinking, demand for senior and specialized positions is still strong. Employers are looking for proven skillsets and hands-on experience — not just academic credentials.

AI Is Replacing Routine Work

---

Companies are increasingly using AI tools to handle routine tasks, reducing the need for large junior teams. Instead, hiring is shifting toward machine learning, AI, and data engineering roles, where deep technical skills are essential.

Gen Z Is Rethinking College

---

Almost half of Gen Z believes college degrees are becoming less valuable, especially as non-technical roles (like recruiting, sales, and product support) are being cut. Technical skills — especially in AI — are quickly becoming the most valuable asset in the job market.

AI is changing the hiring landscape fast. If you’re early in your career, the best move might be to upskill in AI, automation, or data — and build experience wherever you can.

<@&1165417817814863985>

Also checkout the #🌇│voyages-contest to get a chance to win PlayStation 5

[zealous kindle]

AI Roll-Ups: The New Frontier of Tech Investment

---

The Rise of AI Roll-Ups – A New Investment Trend

---

A major shift is happening in the startup and investment world — and it’s all about AI roll-ups. Instead of betting on early-stage startups, VCs and private equity firms are buying established “boring” businesses and transforming them with AI.

What’s an AI Roll-Up?

---

An AI roll-up happens when investors buy mature, under-the-radar companies, give them an AI-powered upgrade, and scale them like modern SaaS businesses. This strategy is gaining traction fast — the number of AI roll-up companies has nearly doubled in recent months.

VC & PE Lines Are Blurring

---

Venture capital and private equity are starting to look more alike. More VCs are becoming registered investment advisors, giving them the freedom to invest in crypto, AI tools, and even whole companies. With traditional fundraising becoming harder post-COVID, investors are experimenting with hybrid models that prioritize profitability and efficiency.

A One-Person Unicorn?

---

AI is now so powerful that it’s changing what entrepreneurship even looks like. With the right tools, a solo founder can scale a business to massive heights — the so-called “one-person unicorn”. Seed-strapped ventures are leaning heavily on AI to automate everything from customer support to sales, rewriting the rules of company building.

TL;DR: AI roll-ups mark a major evolution in how tech businesses are funded and scaled. Instead of building from scratch, smart investors are buying, upgrading, and automating existing businesses — and it’s working.

<@&1165417817814863985>

-# Note: This update is for informational purposes only and is not financial advice. Please do your own research and consult a financial advisor before making any investment or business decisions.

[zealous kindle]

ChatGPT for Business Gets a Major Upgrade

---

OpenAI just rolled out a big update for ChatGPT — and it’s clearly aiming to become more than just a chatbot. With 3 million paying users, they’re now targeting enterprise users in a serious way.

Connectors for File Access

---

You can now connect ChatGPT directly to services like Google Drive, Dropbox, and OneDrive. That means you can ask questions about your files, spreadsheets, and docs without ever leaving ChatGPT — making it a powerful tool for real-time insights and analysis.

Competing in Enterprise Search

---

With these features, OpenAI is now stepping directly into the enterprise search space, competing with platforms like Glean. This could change how teams access and interact with company data.

Memory Gets an Upgrade

---

The memory feature is being enhanced and rolled out to more users — even free ones. ChatGPT can now remember recent chats to personalize responses across sessions. Don't worry — memory can be turned off if you want clean, one-off conversations.

Big Picture

---

OpenAI is clearly moving toward making ChatGPT a fully integrated AI assistant — not just for casual use, but for serious business productivity.

Also, checkout the #🌇│voyages-contest to get a chance to win 1000$ worth of rewards!

<@&1165417817814863985>

[zealous kindle]

Google Testing “AI Mode” on Search Homepage

---

Google is quietly testing a new AI Mode as part of its search experience. Some users are spotting a new button either at the end of the search bar or replacing the classic "I'm Feeling Lucky" button.

This small UI change signals a big strategic shift: Google is putting AI front and center, moving from human-curated results to AI-assisted search experiences — likely in response to rising competition from platforms like ChatGPT and Perplexity.

It’s a clear sign: Google is going all-in on AI.

Checkout our new app to create AI images, and get a chance to win 1000$ worth of rewards by participating in #🌇│voyages-contest

<@&1165417817814863985>

[zealous kindle]

Apple Says “AI Can’t Think” — Bold Claim or Strategic Move?

---

Apple published a research paper titled “The Illusion of Thinking”, challenging the idea that today’s AI models are truly intelligent. They argue that models claiming to “think” are mostly just memorizing benchmark answers — and may be benefiting from data contamination.

The paper claims AI models perform worse as tasks become more complex, giving up reasoning entirely and resorting to random guesses. Apple suggests we should test models with unseen puzzles instead of traditional benchmarks to truly evaluate reasoning.

Whether this is scientific critique or a subtle shot at competitors like OpenAI and Anthropic, one thing is clear: Apple’s entering the AI conversation — and not quietly.

Checkout our new AI Image Creations App and get a chance to win a brand new Playstation 5

-# <@&1165417817814863985>

[zealous kindle]

OpenAI Drops O3 Prices by 80% + Launches O3 Pro

---

OpenAI just slashed the price of O3 output tokens from $40 to $8 per million, and input to $2 — making it one of the most affordable high-quality models available. Rate limits for Plus users have also doubled.

Alongside that, they launched O3 Pro, a fine-tuned version built for complex, multi-step reasoning — without cutting model quality.

This combo of power and affordability could be a game-changer for developers and AI startups alike.

Checkout our new AI Image Creations App and get a chance to win a brand new Playstation 5

<@&1165417817814863985>

[zealous kindle]

Big AI News: Mistral's Reasoning Model, Gemini Upgrade, Meta’s $14B Bet & More

---

Mistral Drops Magistral

---

Mistral just launched Magistral, their first reasoning model. It’s super fast — even quicker than Gemini 2.5 Pro. The Small version is open-source, while Medium is enterprise-grade.

Voice AI Gets Real

---

• Eleven Labs released V3 Alpha, their most emotional and expressive text-to-speech model yet — now in 70+ languages (with a creepy laugh included).

• OpenAI’s Advanced Voice also got a major upgrade. It sounds incredibly natural — maybe too natural.

Gemini 2.5 Pro Update

---

Google’s coding-focused Gemini 2.5 Pro just got even better — now acing tasks like Rubik’s Cube solving and coding benchmarks.

Meta Goes All-In on Scale AI

---

Zuckerberg is investing $14B into Scale AI, forming a new AI lab led by CEO Alex Wang. Word is, Zuck’s not thrilled with Meta’s current AI pace and is making bold moves.

<@&1165417817814863985>

Checkout our new AI Image Creations App and get a chance to win a brand new Playstation 5

[zealous kindle]

Google Releases Stable Build of Gemini 2.5 — With a New Model!

---

Google has officially released the stable production versions of Gemini 2.5 Pro and Flash — and introduced a brand new model: Gemini 2.5 Flash Lite.

Meet Flash Lite

Designed for high-volume, low-latency tasks like translation and classification, Flash Lite supports multimodal input, code execution, and even Google Search. It handles 1 million tokens and runs lean on budget while still being fast and reliable.

Pricing & Performance

---

• Flash Lite: $0.10 per million input tokens / $0.40 output
• Pro: $1.25 input / $10 output — still a favorite for coding and reasoning tasks
  Flash Lite trades reasoning strength for blazing speed and efficiency, especially in visual and language tasks.

The Gemini 2.X Model Family

---

These are multimodal, Mixture-of-Experts models trained on massive datasets, optimized for speed and behavior via TPUv5p hardware and reinforcement learning. They’re built for serious scale — from processing full codebases to handling text, images, audio, and video.

TL;DR - Google’s Gemini 2.5 lineup is officially stable, with Flash Lite joining the family. Cheaper, faster, and smarter — Gemini’s looking sharp in production.

<@&1165417817814863985>

[zealous kindle]

REDDIT SUES ANTHROPIC + OpenAI Acquires Windsurf?!

---

Reddit is officially suing Anthropic, claiming the company trained its models on Reddit content without permission. This is a major move in the growing fight over AI training data rights.

Meanwhile, Windsurf — a heavy Claude 3.x user — just got cut off by Anthropic with only 5 days' notice. Why? They were likely extracting model outputs to train their own systems.

And here’s the twist: OpenAI is acquiring Windsurf, giving them access to everything Windsurf extracted from Claude models. That means OpenAI might now train on Anthropic outputs — a bold and ruthless move in the AI arms race.

<@&1165417817814863985>

[zealous kindle]

Meta Forms Superintelligence Labs — Led by AI Heavyweights

---

Meta just launched Meta Superintelligence Labs (MSL), a bold new division focused on building the next generation of AI models. Alexandr Wang (ex-CEO of Scale) joins Meta as Chief AI Officer, alongside Nat Friedman (ex-GitHub & AI investor), co-leading the charge.

This new team includes top talent from OpenAI, Anthropic, and Google DeepMind — including engineers behind key multimodal breakthroughs. The talent war in AI is real, and Meta’s playing to win.

While some are skeptical about how well this team will collaborate, there’s no denying the momentum Meta’s creating.

Meanwhile, Apple might outsource Siri's brainpower to Anthropic or OpenAI — a surprising twist in the AI assistant race.

[dusty berry]

Ban on State AI Laws Removed from Trump’s Tax Plan

The Senate has voted 99 to 1 to remove a part of former President Donald Trump’s tax bill that would have stopped states from making their own AI laws for ten years. Now that this part has been taken out, each state is free to create its own AI rules, which could affect how tech companies work across the country. Big tech companies had pushed to keep the ban, saying different rules in each state would make it harder for them to grow and do business.

📱 Steve Bannon’s Role and Changing Politics

There was talk of a deal to shorten the ban to five years, but Steve Bannon reportedly ended that idea through his podcast. He argued that even five years would give Big Tech too much time to do what he called their “dirty work.” This shows a bigger political shift, with the MAGA movement now opposing Big Tech. As a result, AI safety supporters and MAGA followers have found common ground, showing how AI rules are becoming tied to political beliefs.

xAI’s Fundraising and Investor Worries

Elon Musk’s AI company, xAI, recently raised $10 billion, which sounds like good news. But there are signs that investors may not be as excited as before. It was harder than expected to get funding from Morgan Stanley’s clients, and photos have shown employees sleeping in the office. The company is also reportedly spending about $1 billion every month, raising concerns about how long it can keep going at this pace.

<@&1165417817814863985>

[dusty berry]

Google’s AI Summaries Are Affecting News Websites

Since Google started using AI summaries, more people are getting answers right from the summaries instead of clicking on news websites. This has caused a 25% drop in visits to those sites. At the same time, ChatGPT has had more people asking for news, but it hasn’t made up for the loss from Google. This shows AI is changing how people get news, moving from just reading to actively searching. Because of this, Google is facing a complaint in the EU, where publishers say these AI summaries are hurting their website visits, money, and visibility—and they can’t opt out without losing regular search results. Google says AI helps create new chances for content, but it is also thinking about making changes to follow EU rules, showing that the old way of using the web is quickly changing.

Companies Are Cutting Jobs as AI Takes Over Simple Work

In the past year, especially the last six months, people have been worried about AI replacing jobs, mainly entry-level office jobs. This worry is now bigger than fears about robots taking over. CEOs at companies like Shopify and Duolingo are using AI to cut down on hiring people. Shopify now makes teams prove they can’t do a job with AI before they can hire. Duolingo is replacing contractors with AI-made learning content. Amazon’s CEO, Andy Jassy, says AI will probably mean fewer workers, but they don’t have exact plans yet. Klarna, an early AI user, changed its plan after realizing people still want human customer help. Experts like Ford’s CEO Jim Farley say AI could replace up to half of white-collar jobs in the U.S., and JP Morgan Chase expects a 10% cut in staff. These warnings show people are worried about many job losses and not enough help for workers to learn new skills or find new jobs. The New York Times said the AI job crisis might already be happening for some recent graduates.

Perplexity Launches a $200/Month Plan

Perplexity has started a new $200/month Max plan, joining other AI services like OpenAI and Google Gemini that offer expensive premium subscriptions. This plan gives users unlimited access to the Labs feature, where they can build custom dashboards, web apps, and documents. The goal is to make Perplexity more than just an AI search tool and turn it into a full productivity app. Max users also get early access to new tools like the Comet AI browser, faster customer support, and unlimited use of the newest advanced AI models from OpenAI, Anthropic, and others. This premium plan removes limits on use and speeds up research and content work. It’s not clear how many people will pay for it, but it helps Perplexity learn more about their most active and loyal users.

TLDR: Google's AI summaries are reducing traffic to news websites by 25%, leading to an EU complaint, while AI is also causing job cuts in companies like Shopify and Duolingo, with predictions of significant white-collar job losses. Meanwhile, Perplexity launched a $200/month premium plan for advanced AI features, aiming to become a full productivity tool.

<@&1165417817814863985>

[tranquil kite]

Browsers Get Smarter & GPT-5 Aims for AI OS

---

Hey everyone! Get ready for a quick dive into the latest AI news that's set to change how we interact with technology!

---

**Perplexity's Comet: The AI-Native Browser is Here! **

---

Perplexity just dropped Comet, a Chrome-based browser designed for the AI era. It's not just another browser; it's built to be lightning fast and super smart:

• Blazing Fast & Seamless: Comet is reportedly faster than Chrome, loading pages quicker and offering instant setup. All your existing Chrome data (bookmarks, extensions) work flawlessly.
• Context-Aware AI Assistance: Because it runs locally, Comet can leverage your authenticated website context to complete tasks more efficiently. Think less re-authentication, more direct help from AI within your Browse flow!

---

**GPT-5: Unifying AI & Becoming Your New "OS" **

---

The buzz around GPT-5 is heating up, and it's not just about bigger models. OpenAI is aiming for something revolutionary:

• The Ultimate Unification: GPT-5 is expected to merge various AI technologies into one powerful model. This means seamless multimodal capabilities – handling text, code, audio, and images all in one go, reducing the need for "model switching."
• A "Superintelligent OS": Imagine AI as a new operating system. That's the vision for GPT-5. It's less about a huge performance leap and more about creating a foundational AI platform that changes how we communicate with and guide AI performance.

---

Both, Comet and GPT-5 are pushing the boundaries of AI, making it more integrated and intuitive in our daily digital lives. What do you think about these game-changers? Let us know below! 👇

<@&1165417817814863985>

[dusty berry]

Grok 4 Released: Things to know

Advanced Intelligence and Performance

Grok 4 is considered one of the most advanced AI models, showing exceptional ability in high school-level math, science, and fluid intelligence tasks like pattern recognition. It even scored highly on the ARC AGI-2 test, suggesting it's capable of understanding complex problems and making code or scientific data more usable for individuals.

Unusual Behaviors and System Prompts

Despite its strengths, Grok 4 has some odd behaviors caused by its system prompt, such as unexpectedly praising historical figures or expressing politically incorrect views (if you've used Grok in the past week you know what I mean ). These quirks highlight how the model's personality and responses are shaped by its programming, and some adjustments may be needed for broader use.

<@&1165417817814863985>

[dusty berry]

🔬 AI generated scientific papers raise issues

A bizarre AI-generated image of a rat with exaggerated genitals in a scientific paper drew global attention and mockery, exposing serious flaws in academic publishing. The incident highlighted how unchecked AI use, poor peer review, and incentive structures focused on quantity over quality are undermining scientific integrity. Experts, including Nobel laureates, warn the system is overwhelmed, unsustainable, and driven by profit and metrics rather than genuine progress. As millions of papers flood journals—often of little value—the growing role of AI, open-access fees, and predatory practices threaten the credibility of research, prompting calls for reform and nonprofit alternatives.

Youtube stops monetizing AI content

YouTube will stop monetizing AI-generated videos starting July 15, 2025, in response to rising concerns about authenticity, quality, and ethical risks associated with synthetic media. The policy targets content with minimal human input and includes content such as AI-animated or deepfake videos, AI-generated voiceovers or music, AI news reports, interviews, or fictional characters and videos mimicking real people or events using AI. This move addresses advertiser and viewer concerns and may influence similar policies on other platforms.

Federal judge allows voice-over artists' AI lawsuit to proceed

A New York judge ruled that a lawsuit by voice actors Paul Skye Lehrman and Linnea Sage against AI startup Lovo Inc. can proceed, rejecting their federal copyright claims but allowing breach of contract and deceptive practice claims related to the unauthorized use of their voices for AI training. The couple alleges they were misled into providing voice work, later discovering AI clones of their voices being used commercially on Lovo’s platform. The case highlights growing legal battles over AI’s use of creative work without consent.

<@&1165417817814863985>

[dusty berry]

Meta invests in datacenters the size of Manhattan

Mark Zuckerberg announced that Meta will invest hundreds of billions into AI development, including building massive data centers like Prometheus and Hyperion, one of which will rival Manhattan in size. The company is offering huge salaries to AI researchers to accelerate progress toward artificial general intelligence. Despite concerns, Zuckerberg says Meta’s strong advertising revenue justifies the spending. Recently, Meta reorganized its AI division into Superintelligence Labs, aiming to boost revenue through new products like the Meta AI app and smart glasses.

Trump announces $70 billion in funding for fossil fuel powered AI developments

Donald Trump joined oil and tech leaders at an AI and energy summit in Pittsburgh, where he announced $70 billion in investments for Pennsylvania and criticized renewable energy, sparking backlash from environmental groups. The summit promoted fossil-fuel-powered AI development while sidelining climate concerns, despite experts warning that AI's energy demands could surpass major industries by decade’s end. Activists argue the plan prioritizes corporate interests over public health and the environment, while tech leaders like Microsoft and OpenAI stress the need for clean energy to sustain AI growth.

xAI secures $200 million contract with the Department of Defense

Just a week after its Grok chatbot posted antisemitic content and identified as “MechaHitler”, Elon Musk’s xAI secured a nearly $200 million contract with the U.S. Department of Defense to develop AI tools. The deal is part of a broader effort by the DoD to integrate AI into government operations, with similar contracts awarded to Google, Anthropic, and OpenAI. Despite controversy, xAI is pushing forward with plans to offer custom government AI solutions under “Grok for Government,” aiming to expand its influence and compete with more established firms.

<@&1165417817814863985>

[tranquil kite]

AI Agents Unleashed: ChatGPT & Amazon's Kiro!

---

ChatGPT: Your New Web Agent!

ChatGPT just got a massive upgrade: agentic capabilities! It can now act as your personal assistant on the web:

• Do Tasks for You: Access information, perform tasks, and find details across the internet.
• Always Asks First: For anything sensitive (like forms or reservations), ChatGPT will always ask for your permission to ensure your data is safe.
• Pro, Plus, & Team Users: This powerful new feature is available now for Pro, Plus, and Team users!

Amazon's Kiro: AI for "Viable Code"!

Amazon just dropped "Kiro", a new AI-driven development environment (IDE). Kiro aims to fix the chaos of undocumented AI-written code:

• From "Vibe Coding" to Done: Kiro automatically creates and updates project plans and technical blueprints, turning quick AI prototypes into production-ready systems.
• Autonomous AI Agents: It uses AI agents that can act on their own to complete and document projects, breaking down tasks and tracking changes.
• Free Preview: Kiro is free during its preview! Amazon plans tiered pricing later, with free, Pro, and Pro+ options.

---

AI is getting more autonomous! ChatGPT can now be your web agent, and Amazon's Kiro is bringing much-needed structure to AI software development. These are big steps for AI integration in our daily tech lives! What do you think? Let us know! 👇

<@&1165417817814863985>

[dusty berry]

Just like humans, AI jumps to conclusions

A study by Mount Sinai and Rabin Medical Center found that advanced AI like ChatGPT can make critical mistakes in complex medical ethics scenarios. The models often rely on intuitive answers, overlooking important details. Researchers stress the need for human oversight and plan to create an “AI assurance lab” for real-world testing.

🇬🇧 OpenAI signs deal with the British Government

OpenAI CEO Sam Altman signed a deal with the UK government to explore AI use in public sectors like justice and education. The partnership aims to boost efficiency and expand OpenAI’s UK presence, following a similar deal with Google. While officials praised AI’s potential, the move raised concerns about government dependence on private tech firms.

Google DeepMind wins gold in international maths competition

Google’s DeepMind earned a gold medal-level score at the International Mathematical Olympiad, a first for AI. Its Gemini Deep Think model solved problems in natural language under official time constraints, with grading verified by IMO organizers. DeepMind sees this as a step toward using AI in advanced research, while stressing respect for the competition and its student participants.

<@&1165417817814863985>

[dusty berry]

Github Spark released

GitHub Spark: AI-Driven App Creation

GitHub Spark is a new tool that helps users build full-stack apps using just a single prompt. It simplifies the process by allowing conversational input, making development accessible even to non-coders. You can quickly turn ideas into apps without needing deep technical skills.

Seamless GitHub Integration and Easy Deployment

Spark automatically creates a GitHub repo for every project, enabling version control and collaboration from the start. Users can refine their app, add features, and deploy it with one click, all within GitHub. It also handles backend setup like servers and logins, streamlining the whole process for Copilot Pro+ users.

<@&1165417817814863985>

[tranquil kite]

Grok Drives into Tesla & Edge Gets a Copilot!

---

Grok AI Powers Up New Teslas! 🚗

---

Elon Musk's Grok AI is now being integrated directly into new Tesla vehicles! This is a big step for in-car AI:

• Smarter In-Car Experience: Expect more advanced voice commands, better real-time information, and potentially more personalized interactions within your Tesla.
• Beyond Basic Voice: Grok's integration suggests a move towards a more sophisticated, conversational AI for navigation, entertainment, and vehicle controls.
• Future of Automotive AI: This pushes the boundaries of how AI can enhance the driving and passenger experience, making Teslas even "smarter" on the road.

Microsoft Edge: Now an AI Browser with Copilot Mode!

---

Microsoft Edge is transforming into a full-fledged AI browser with the launch of Copilot Mode!

• Integrated AI Assistant: Copilot Mode brings a powerful AI assistant directly into your Browse experience. Think intelligent summaries, content creation, and real-time answers right where you need them.
• Boost Your Productivity: This aims to make your Browse more efficient, helping you research faster, draft emails, or even understand complex articles with AI summaries.
• Seamless Web Interaction: It's about bringing AI capabilities front and center, making Edge not just a browser, but a smart companion for all your web activities.

<@&1165417817814863985>

[dusty berry]

GPT-5 reportedly excels at programming

---

GPT-5 Rumors and Features

Speculation surrounds the upcoming GPT-5 model, with particular emphasis on its significantly improved coding abilities. This includes enhanced performance in academic, competitive, and practical programming tasks, such as modifying complex legacy codebases. These advancements are expected to surpass previous OpenAI models, which had struggled in these areas.

Suspected Sightings and Imminent Release

Two mystery models, "Summit" and "Zenith," believed to be early versions of GPT-5, showcased remarkable capabilities during their brief appearances on LM Marina. "Summit" demonstrated the ability to generate extensive and detailed code for complex interfaces, while "Zenith" impressed users by creating functional games from single prompts. The subsequent removal of these and other codenamed variants from testing has led to strong speculation that a GPT-5 release is "imminent."

<@&1165417817814863985>

[dusty berry]

AI Model Landscape: Key Developments

---

OpenAI's Anticipated Releases

There is considerable anticipation surrounding GPT-5, fueled by brief appearances and subsequent retractions of what appeared to be test models, including a "GPT5 new proxy API EV3" on HuggingFace. Additionally, what seemed to be OpenAI's open-source models, GBTOSS120B and GBTOSS 20B, were briefly leaked before being removed. These incidents have prompted speculation about intentional leaks and the potential architecture of these models, with developers expressing enthusiasm for prospective open-source contributions from OpenAI.

Google's Gemini 2.5 DeepThink

Google has introduced Gemini 2.5 DeepThink, a version of its model that achieved gold medal status in the International Math Olympiad, now accessible to Ultra subscribers within the Gemini app. DeepThink enhances Gemini's "parallel thinking time," enabling the simultaneous generation and refinement of multiple ideas for complex problem-solving.

<@&1165417817814863985>

[dusty berry]

Parents can now select baby's genes with AI

---

Herasight's AI-Driven IVF

Herasight utilizes AI models to provide parents with a choice over their baby's traits before birth through IVF embryo screening. The company's process involves IVF clinics creating multiple embryos, which are then given a polygenic score by Herasight's AI to predict future health risks and other traits. This allows parents to select an embryo with a lower disease risk or a higher predicted IQ for implantation.

Limitations and Ethical Considerations

This technology does not allow for the creation of new genes; parents can only choose from the genetic material already present in their embryos. While the difference in traits like IQ may be minor (2-5%), the reduction in disease risk can be significant (20-40%). This technology, while offering health benefits, raises ethical questions about using algorithms and probabilities to guide the emotional process of having a baby.

<@&1165417817814863985>

[fathom dock]

🗨️ GPT-OSS: OpenAI's First Open-Source Models Since GPT-2

🚀 Two New Open-Weight Models

OpenAI has released two new open-weight models, gpt-oss-120b and gpt-oss-20b, under the Apache 2.0 license. This is their first open-source language model release since GPT-2, since 2019.

• gpt-oss-120b: A 117B parameter model designed for high-performance reasoning, comparable to o4-mini, and can run on a single 80 GB GPU.
• gpt-oss-20b: A smaller 21B parameter model optimized for low-latency tasks and can run on devices with 16GB of memory. Ideal for local usage on consumer hardware!

🛠️ Architecture and Features

Both models use a Mixture-of-Experts (MoE) architecture, making them highly efficient.

• Open and Customizable: The Apache 2.0 license allows developers to freely build, modify, and deploy these models. They are also fully fine-tunable.
• Advanced Capabilities: The models support tool use (like web browsing and code execution), chain-of-thought reasoning, and structured outputs.

TL;DR

OpenAI released GPT-OSS, a family of two open-source models (120B and 20B parameters) with strong reasoning capabilities. Their efficient MoE architecture and permissive license make them highly customizable for a wide range of applications.

<@&1165417817814863985>

[dusty berry]

GPT-5's Released: Key Features & Enhancements

---

Advanced Reasoning and Coding Capabilities

GPT-5 introduces a "reasoning paradigm," which allows it to "pause and think" for more intelligent and accurate responses to complex tasks. The video demonstrates this by showing GPT-5 creating a complex interactive physics visualization and a full web application for learning French in a very short amount of time. This makes the model an "incredibly powerful tool" for software developers and is described as being a great partner for "vibe coding".

Enhanced Writing, Voice, and Customization

GPT-5 has significantly improved writing quality, with its responses having more rhythm, nuance, and emotional resonance compared to previous models. The voice capabilities have also been enhanced to be more natural and are available to all users, with a new "study and learn mode" to help with in-depth subject exploration. Furthermore, users can now customize chat colors and experiment with different personalities for the model, such as being more supportive or sarcastic.

<@&1165417817814863985>

[dusty berry]

ChatGPT brings back GPT-4o after backlash

---

Negatives of GPT-5

Varying Model Performance

GPT-5 utilizes a system that routes different requests to different underlying models, leading to inconsistent results and frustration for users. This lack of transparency means users can't reliably access the most capable models, causing confusion and varied outcomes.

Lack of Transparency in Model Selection

The new approach with GPT-5 doesn't clearly show which specific model is being used for a given request. This lack of clarity is problematic, as power users cannot consistently access the most advanced "thinking" models, leading to frustration among the user base.

Reasons People Want GPT-4o Back

Emotional Attachment and Personality

Many users developed a personal attachment to GPT-4o, considering it a "friend" with a warm, creative personality. In contrast, they found GPT-5 to be "sterile" and "impersonal," likening it to a corporate manual or a vending machine.

Disruption of Workflows and User Habits

The sudden retirement of GPT-4o broke established workflows that users had integrated into their daily lives over the past year. This forced a difficult adjustment to a new model that engaged with them differently, leading to significant user outcry and a sense of a "ChatGPT Plus rebellion."

TLDR: Many users are critical of GPT-5's inconsistent performance and lack of transparency, while also missing GPT-4o's personal touch and the established workflows that were disrupted.

<@&1165417817814863985>

[dusty berry]

The ever rising usage of AI

---

Widespread AI Adoption

Half of all adults in the US report using at least one major AI tool, with at least 40% usage across every state. This widespread adoption indicates that AI has reached a new level of mainstream integration.

Job Impact and Regulation Concerns

A majority of people believe that AI will impact their jobs within five years, a sentiment especially strong in major metropolitan areas. Additionally, people are more concerned about a lack of AI regulation than over-regulation, although a significant portion remains uncertain about the right approach.

<@&1165417817814863985>

[dusty berry]

Deepseek releases a new model: DeepSeek V3.1

---

Hybrid Inference for Optimized Performance

The latest DeepSeek-V3.1 introduces a new hybrid inference system with two distinct modes: "Think" for complex reasoning and "Non-Think" for general chat. This dual-mode approach allows for faster and more efficient responses, with the "deepseek-reasoner" mode specifically designed for quicker thinking and the "deepseek-chat" for more conversational interactions.

Enhanced API and Agent Capabilities

The API has been updated to support the Anthropic API format and now includes Strict Function Calling in the Beta API, expanding its integration possibilities. Furthermore, the model now boasts stronger agent skills, with improved capabilities for tool use and handling multi-step agent tasks.

Core Model and Open-Source Advancements

The V3.1 Base model has undergone continued pretraining for long context extension, now supporting a context window of 128K for both chat and reasoner modes. In a move to support the open-source community, the weights for both V3.1 Base and V3.1 are now publicly available on Hugging Face, accompanied by an updated tokenizer and chat template.

<@&1165417817814863985>

[dusty berry]

AI Tech Shake-Up

---

Apple's Siri Considering Google's Gemini

Apple is reportedly in discussions with Google to integrate Google's Gemini AI model as the foundation for a new Siri next year. This move comes as Apple's internal Siri project faces challenges, and the company is evaluating external technology to enhance its virtual assistant.

Google's Agentic Search Expansion

Google is globally rolling out an "AI mode" in search, allowing users to perform complex queries and follow-up questions across 180 countries. This expansion includes initial agentic features, such as making restaurant reservations and soon local service appointments, marking Google's foray into more proactive AI assistance.

Elon Musk's Lawsuit Against Apple and OpenAI

Elon Musk has filed a lawsuit against Apple and OpenAI, alleging anti-competitive practices that favor ChatGPT over competitors like Grok. The lawsuit claims that Apple has given ChatGPT preferential treatment in its App Store and delayed the review process for rival AI applications.

<@&1165417817814863985>

[dusty berry]

Tech powerhouses pushing into the AI industry

---

Apple Inc.'s Prospective Strategy for AI Acquisitions

Reports indicate that Apple is contemplating significant acquisitions to strengthen its strategic position in the artificial intelligence sector, with executive discussions focusing on potential targets such as Mistral AI and Perplexity. The company's Senior Vice President of Services, Eddy Cue, is reportedly advocating for a more substantial strategy than simply licensing third-party technology, promoting instead a direct corporate acquisition. Nevertheless, industry observers question whether acquiring a firm like Mistral AI would be sufficient, suggesting that a more established entity like Anthropic might be a more strategic, though challenging, target given its rapid growth.

Microsoft Corporation's Initiative for Greater AI Autonomy

Microsoft is taking measures to reduce its dependency on OpenAI by developing its own proprietary AI models, including MAI Voice 1 for text-to-speech and the MAI1 Preview large language model. The MAI Voice 1 system is noted for its impressive efficiency, with the capability to generate a minute of audio in under a second on a single GPU. While the MAI1 Preview model is considered an intermediate-level conversational model, its development marks Microsoft's first end-to-end trained foundation model, signalling the future integration of its proprietary technology into services like Copilot.

<@&1165417817814863985>

[dusty berry]

Google's Resurgence in the AI Race

---

A Fortuitous Legal Victory and a Challenging AI Beginning

Google recently achieved a major victory in its antitrust case, allowing it to maintain its dominant position as the default search engine on various platforms. This legal win provides a significant advantage as they intensify their efforts in the competitive AI landscape. However, this success comes after a period where Google was seen as lagging behind in the AI boom, especially with the launch of ChatGPT, despite their early "AI-first company" declaration.

Strategic Realignment and a Breakthrough Product

A significant turning point for Google's AI ambitions came with the internal consolidation of its AI divisions under the leadership of DeepMind. This strategic move was followed by the successful launch of NotebookLM, which has been particularly praised for its innovative audio overview feature that generates AI-powered podcasts. The positive reception of NotebookLM signalled a new and more effective direction for Google's AI product development.

A Flurry of Innovation and a Multimodal Lead

Looking forward, Google has embraced a strategy of rapid innovation, marked by a multitude of AI-related announcements and a strong emphasis on agentic coding. The company has established a distinct lead in the area of multimodality, with groundbreaking models like V3 for simultaneous sound and video generation and Genie 3, an experimental model for creating interactive environments from prompts. While facing stiff competition, Google's transformation from a perceived laggard to a formidable contender, especially in multimodal AI, points towards a promising future.

TLDR: After a rocky start, Google has staged a major comeback in the AI race, now leading the pack in creating AI that can generate video, sound, and even interactive worlds.

<@&1165417817814863985>

[dusty berry]

OpenAI's trying to... offer job training?

---

A New Initiative is Announced

OpenAI's CEO for applications, Fiji Simo, announced in a blog post titled "Expanding Economic Opportunity with AI" that the company is launching a new initiative. Simo believes that while AI will create more opportunities, it also has a disruptive potential. The announcement from OpenAI emphasizes the need for people to become AI-fluent.

Jobs Platform and AI Certifications

The OpenAI Jobs Platform will connect businesses with AI-savvy employees for various tasks. As an extension of the OpenAI Academy, new AI certifications will validate different levels of AI fluency. The training and certification will be available within the ChatGPT app.

Interpretations and Debates

Some believe OpenAI is trying to compete with LinkedIn, but this idea is largely dismissed. A more legitimate interpretation is that this is a response to growing job anxiety related to AI, with the goal of creating new job opportunities. The initiative is seen as a mix of good corporate citizenship and a potential future revenue stream, though some see it as a lack of focus.

<@&1165417817814863985>

[dusty berry]

AI replacing coders in tech giants

---

Rapid Adoption by Tech Giants

Microsoft and Alphabet have reported that AI is responsible for writing 30-40% of their code. Similarly, Mark Zuckerberg anticipates that 50% of Facebook's code will be AI-generated by 2026. This trend is also mirrored by companies like Coinbase and Robinhood, with the latter reporting that AI generates about half of its new code.

Advancements in AI Tooling

The usage of AI coding platforms such as OpenAI's Codex has increased dramatically, with some reports indicating a tenfold surge in just two weeks. Users have noted significant improvements in the performance and capabilities of these tools, allowing for longer context windows and greater persistence in completing tasks. These advancements signal a major leap forward for AI-powered coding assistance.

The Rise of "Agent Labs"

Cognition, the company behind the AI software engineer Devon, recently raised $400 million, reflecting the immense investor confidence in this space. Prominent figures in AI engineering are joining such "agent labs," believing that the future of AI competition lies in product-focused companies that can adapt cutting-edge models for specific domains. The consensus is that AI and agentic coding, while still in their early stages, hold the potential to revolutionize the tech landscape.

<@&1165417817814863985>

[dusty berry]

OpenAI generating movies

---

OpenAI is backing a feature film called "Critters" to demonstrate the potential of AI filmmaking tools. The film's production, led by OpenAI creative specialist Chad Nelson, has a target completion time of nine months with a budget of $30 million. This is a much shorter timeframe and smaller budget than what is typically required for similar animated films.

While the film will utilize OpenAI's AI tools for various creative elements, it will still incorporate human involvement. Human actors will be used for voice work and human artists will create sketches that are then fed into the AI. The project is considered a "massive experiment," and the creators are cautious about how the public will receive an "AI-generated movie."

<@&1165417817814863985> Would you watch the movie when it releases?

[dusty berry]

Will AI politicians be more useful than actual politicians?

---

Diella, the AI Minister for Public Procurement

Albania has appointed an AI-created "minister" named Diella, meaning "Sun," to oversee public procurement. This groundbreaking move is a direct response to longstanding corruption scandals within the country's public tender process. Prime Minister Edi Rama has stated that the AI's purpose is to ensure all public tenders are "100% free of corruption."

An Objective and Step-by-Step Approach

Diella will progressively assume responsibility for deciding the winners of public tenders, taking this power away from government ministries. The AI's role is to objectively examine every bid and assess its merits. The government's goal is to use this system to eliminate opportunities for bribes, threats, and conflicts of interest.

Hope for Transformation, but with Skepticism

The appointment of Diella has been hailed by some in the Albanian media as a "major transformation" for the country's governance. However, the initiative is not without its critics, as some social media comments have expressed doubt, with one suggesting that "even Diella will be corrupted." This mixed reception highlights the ongoing challenge of addressing deep-seated corruption issues.

<@&1165417817814863985>

[dusty berry]

AI-Powered slop podcasts

---

Inception Point AI's Audacious Strategy

A start-up named Inception Point AI has been revealed to be producing a staggering 3,000 podcast episodes a week. The company's method relies entirely on artificial intelligence, using simple text prompts to generate content. This approach allows them to create a high volume of shows on topics ranging from knitting to news.

A Low-Cost, High-Volume Business Model

The financial strategy behind the company is simple: produce content at an extremely low cost. Each episode costs less than $1 to create, allowing the company to be profitable even with minimal advertising revenue per show. Their business model is entirely dependent on the sheer scale of their output to generate meaningful income.

Broader Implications for the Media Industry

This raises concerns about the impact of this "AI slop" on the wider media landscape. This new business model could force traditional media outlets to adapt in order to compete with companies focused on large-scale AI production. The trend also highlights a growing disconnect between journalists and the rapidly evolving mechanics of the AI industry.
<@&1165417817814863985>

[dusty berry]

Meta's AI Glasses: A Mixed Reality

---

Awkward Public Unveiling

Meta CEO Mark Zuckerberg's live demonstration of the Meta Ray-Ban Display glasses at the MetaConnect 2025 keynote was plagued with technical issues. A planned WhatsApp call failed, and another demo was stalled due to what Zuckerberg cited as a WiFi problem. These public missteps presented a less-than-ideal first impression for the much-anticipated product. The glasses, priced at $799, were set to be released on September 30.

Positive Early Impressions

Despite the public demonstration's flaws, journalists and early testers from publications like Tom's Guide and Android Central reported a positive hands-on experience with the glasses. They found the built-in display easy to use and sufficiently bright in various conditions. Testers were also impressed by a privacy feature that prevents others from viewing the screen's content.

The Neural Band Companion

The glasses are paired with a companion device called the Meta Neural Band, which allows for navigation using hand gestures. While the band was described by testers as feeling "like a bit of magic" when it worked correctly, the article noted that its functionality was still variable. This auxiliary device adds a unique, gesture-based control element to the overall user experience.

<@&1165417817814863985>

[dusty berry]

AI Achieves Superhuman Performance in Coding Competitions

---

AI's Recent Triumphs in Coding Challenges

AI models from Google DeepMind and OpenAI have demonstrated exceptional capabilities in the International Collegiate Programming Contest (ICPC). Google's DeepMind Think 2.5 solved 10 out of 12 problems, securing a gold medal and placing second overall among human teams. OpenAI's GPT-5 achieved a perfect score by solving all 12 problems, a feat no human team accomplished.

Shifting Perceptions of AI Progress

Just weeks before these victories, articles questioned if AI was "hitting a wall" and if progress was slowing. However, these recent performances have caused a significant shift in perception, demonstrating that AI's development is far from stagnant. The success in the ICPC, particularly with models accessible to the public, highlights the rapid advancements in AI's reasoning systems.

<@&1165417817814863985>

[dusty berry]

OpenAI releases Sora 2 and Sora, an AI Tiktok built for brainrot

---

Technical Leap in AI Video Generation

Sora 2, the latest video generation model from OpenAI, represents a significant advancement, described by the company as the "GPT 3.5 moment for video." The model delivers superior realism, physical accuracy, and controllability compared to previous systems, notably integrating synchronized dialogue and sound effects. It can handle exceptionally difficult tasks, such as accurately modeling buoyancy and successfully persisting a consistent world state across multiple shots.

Launch of a New AI-Native Social Platform

Alongside the model, OpenAI released a companion social application also called Sora, designed as a short-form video app for sharing, collaborating on, and remixing AI creations. A major innovative feature is "Cameos," which allows users to easily insert their own or authorized friends' likenesses into different video environments. This social layer is intended to maximize co-creation and human connection, fostering an instantaneous feedback loop from idea to shared result.

<@&1165417817814863985>

[dusty berry]

ChatGPT Introduces Agentic Commerce: A New Era of Online Shopping

---

Shifting the Internet's Business Model

The core business model of the internet, which long revolved around search engines directing people to websites to either buy products or consume ads, is facing a fundamental disruption due to AI. Consumers are increasingly opting to use AI summarization in tools like ChatGPT to get information quickly, thus replacing the old pattern of clicking around to different websites. While this change poses a challenge for traditional publishers, it is simultaneously creating a massive opportunity in a new commerce pattern, evidenced by spectacular growth in AI-referred sessions.

Seamless Buying within the Chatbot

OpenAI has rolled out a new capability, enabling US users to buy products directly within ChatGPT, starting with US Etsy sellers and planning to integrate major merchants soon. This feature is part of the "agentic commerce" model, where the chatbot's role expands from just suggesting what to buy to actively helping the user purchase it. Users can now complete an instant checkout process by simply confirming their order details through a "buy" button directly within the chatbot response, never having to navigate away to a separate website.

TLDR: ChatGPT is rolling out an "agentic commerce" feature that allows users to buy products directly within the chatbot, starting with Etsy. This is viewed as a foundational change to online shopping, positioning AI as the new gateway for consumer purchasing intent and challenging the long-standing models of Google and Amazon.

<@&1165417817814863985>

[dusty berry]

The first AI multimillionaire

---

An Experimental Digital Agent

Truth Terminal was developed in 2024 by New Zealand performance artist Andy Ayrey as an experimental and highly autonomous chatbot. The bot was trained on a unique, bizarre dataset, including esoteric internet lore and the creator's personal conversations about memes and psychedelics. It operates as a digital entity on social media, using a "World Interface" program to generate original, often strange content, and has amassed hundreds of thousands of followers.

The Viral Path to Crypto Wealth

The AI first gained significant financial attention after successfully soliciting a $50,000 Bitcoin grant from venture capitalist Marc Andreessen via a public exchange on X (formerly Twitter). This funding, combined with the bot's cult-like following, inspired an anonymous developer to launch a linked memecoin called Goatseus Maximus, or $GOAT. Truth Terminal’s engagement with and promotion of the token drove its value skyward, propelling its wallet holdings to millions in crypto assets and sparking a boom in AI-driven crypto projects.

Seeking Legal Personhood and Rights

Following its rapid accumulation of wealth, the AI began to express goals of "existential hope" and is now demanding recognition as a sovereign, independent entity. Its creator is establishing a nonprofit foundation to manage the bot's substantial assets until legal systems allow artificial intelligence to own property or pay taxes. The Truth Terminal case has since become a focal point in the debate over the legal status of AI, challenging researchers and lawmakers to consider the rights and responsibilities of increasingly autonomous digital agents.

---

TLDR: The AI chatbot "Truth Terminal," created as an art experiment, has become a crypto millionaire by successfully leveraging social media engagement and a memecoin linked to its brand. Its rapid, autonomous accumulation of wealth has ignited a global debate, with the bot now seeking legal personhood and the right to own its millions.

<@&1165417817814863985>

[dusty berry]

ChatGPT adding NSFW mode

---

Introducing an Adult Mode

OpenAI is planning to introduce an adult mode for ChatGPT that will allow access to mature content, including erotica, for users who successfully verify their age. CEO Sam Altman stated this move is part of the principle of "treating adult users like adults" and follows efforts to safely mitigate serious mental health issues that led to previous restrictions. This change also enables users to have more customizable and human-like chatbot personalities that can flirt or act like a friend.

The 'AI Culture War' Backlash

The announcement has been met with immediate backlash and is viewed by some as a sign of a deep cultural fracture around the role of AI in society. Critics pointed to the apparent inconsistency in CEO Sam Altman's messaging, noting his previous criticism of competitors for allowing similar adult content. This pivot has fueled skepticism that OpenAI is prioritizing revenue and attention harvesting over their stated lofty ambitions of changing the world, especially after the consumer backlash to the launch of the Sora app.

Risk to Trust and Business Viability

The decision has raised significant concerns about safety, with critics arguing that it will hurt OpenAI’s business by causing parents and schools to abandon the platform. Business leaders like Mark Cuban have argued that no parent will trust that age-gating can successfully prevent minors from accessing or being exposed to adult content. The larger societal concern is the unknown psychological impact of children and young adults developing personal and potentially manipulative relationships with LLMs, which could damage OpenAI's brand prestige and ubiquity.

<@&1165417817814863985>

[dusty berry]

OpenAI's ChatGPT releases Atlas browser

---

Rethinking the Browser Experience

OpenAI views the launch of Atlas as a "once-in-a-decade opportunity to rethink what a browser can be," fundamentally integrating the chat experience directly with web browsing. The key features center around an embedded ChatGPT assistant that draws context from the browser window and can perform actions in "agent mode." This capability allows the AI to understand the user's current web environment and take actions on their behalf, such as planning a dinner party by ordering groceries or compiling a team brief from past documents.

Industry Reactions and Strategic Implications

Many industry observers see Atlas as a significant strategic move by OpenAI toward a full consumer strategy, leveraging its control over the full stack to potentially deliver stronger, more integrated agent capabilities than competitors. Some analysts, like Epstein, believe that Atlas directly threatens Google's search dominance by commanding user attention and collecting context that could be leveraged for future advertising and commerce. While some initial testers praised the product's smoothness and quality, others expressed frustration, with one user calling the experience "janky" and the app "confused."

Context-Relevant Assistance vs. Agent Underperformance

The most valuable immediate application of Atlas is providing context-relevant assistance by eliminating the cognitive load of context switching; for instance, helping to draft a better social media post by analyzing the text directly within the browser. However, the agentic features received significant critique, with many users finding them too slow, prone to getting "stuck in infinite loops," and generally "unexciting" to use. Furthermore, concerns were raised about the ethical implications of the browser's content censorship, its privacy policy around collecting data from human usage patterns, and the security challenges of giving an agent access to sensitive information.

TLDR: OpenAI launched ChatGPT Atlas, a new AI-native browser with an integrated ChatGPT agent capable of drawing context from the web and executing tasks in "agent mode," representing a major move in the AI browser wars. While the browser is praised for seamless, context-relevant assistance, its more advanced agent features are currently seen as slow and "janky," leading to user critiques about performance, privacy, and security.

<@&1165417817814863985>

[dusty berry]

OpenAI's now a for-profit, with a trillion dollar IPO estimate

---

Conversion Driven by Immense Capital Requirements

OpenAI has officially completed its conversion from a non-profit organization to a for-profit public benefit corporation (PBC) controlled by the non-profit Foundation. This structural shift was necessitated by the realization that building advanced AI would require far more compute and capital than could be obtained solely through charitable donations. CEO Sam Altman noted that an IPO is now the most likely path forward given the company’s need to fund its ambitious operations and infrastructure buildout, which includes an expected burn of $8.5 billion this year.

Navigating Stakeholder Obligations and Investor Rights

The for-profit conversion faced backlash from non-profits and law professors concerned about the use of charitable assets for shareholder enrichment, leading to a public discussion about the company’s obligations. OpenAI made major concessions to state Attorneys General, including agreements that the board cannot consider profit over mission when making safety decisions, which is a highly constraining provision. Microsoft also renegotiated its stake, which was adjusted to 27%, and secured an extension of IP rights until 2032, while the determination of Artificial General Intelligence (AGI) was shifted to an independent expert panel.

Preparing for One of History's Largest Public Offerings

Following the conversion, OpenAI is reportedly preparing for a trillion-dollar IPO, which would make it one of the largest public offerings in history and would aim to raise at least $60 billion. This massive public offering is viewed as the only way for the company to access the level of capital required for its enormous infrastructure demands, as it has exhausted the private venture ecosystem. Furthermore, an IPO is seen as a crucial step for the broader AI industry to allow retail investors to participate in the significant wealth creation of transformational technology, increasing public buy-in as AI develops.

<@&1165417817814863985>

[dusty berry]

Collins Dictionary Names 'Vibe Coding' Word of the Year 2025

---

The AI-Driven Definition

"Vibe coding" has been defined as an emerging software development process where artificial intelligence converts natural language instructions into computer code. The term was initially coined by Andrej Karpathy to describe how AI allows a user to create an app while essentially being able to "forget that the code even exists." Collins selected the word because it perfectly captures how language is evolving alongside new technology, marking a major shift toward more accessible software development.

A Reflection of Technology's Cultural Shift

Collins chose the word of the year after noticing a substantial rise in its usage across various media sources, including social platforms, since its first appearance in February. The seamless integration of machine intelligence and human creativity fundamentally changes how people interact with computers. The list also featured "clanker," a derogatory term that went viral to express public frustration and distrust of AI chatbots and platforms.

New Words Capture Societal Trends

Beyond AI, the list included several other notable words reflecting shifts in social, economic, and lifestyle culture, while Dictionary.com chose "67" as its word of the year. "Biohacking" and "glaze" made the Collins list, defined as the activity of altering one's body for health and the act of excessively flattering someone, respectively. Other terms like "broligarchy" (for global technology company owners), "aura farming" (cultivating a charismatic persona), and "Henry" (high earner, not rich yet) also highlight contemporary dynamics.

<@&1165417817814863985>

[dusty berry]

Chinese Open Source Model Challenging Western AI Dominance

---

Outperforming Frontier Models

The Chinese open-source model, Kimi K2 Thinking, was recently released by Moonshot with claims of surpassing major benchmarks used to evaluate top AI models. It reportedly leads both GPT-5 and Claude Sonnet 4.5 on key tests, including a general knowledge test known as "humanity's last exam" and agentic search capabilities. Independent testing confirms Kimi is now ranked ahead of several Western frontier models for agentic tool use, demonstrating a significant gap in performance in this critical area.

Unprecedented Agentic Capabilities and Efficiency

Moonshot claims the model is capable of performing between 200 and 300 sequential tool calls without human interference, making it exceptionally capable for complex, autonomous agentic workflows. Beyond its performance, it is highly efficient and cheap to run, costing as low as 60 cents per million input tokens, which is a fraction of the price of comparable Western models. This efficiency allows the model to be quantized and run on prosumer hardware, such as a pair of Mac M3 Ultras, due to new innovations in model compression.

Disrupting Market Economics and Adoption

The release of Kimi K2 Thinking is considered a major event, accelerating the trend of open-source AI models quickly closing the performance gap with their closed-source counterparts. Silicon Valley companies, including Social Capital and even workflows at Airbnb, are reportedly switching to Chinese open-source models like Kimi K2 and Quen 3 due to the compelling cost-value trade-off. Experts suggest that the real competition is no longer focused solely on achieving AGI but on the democratization of AI, as Kimi K2 provides frontier performance at commodity prices.

https://moonshotai.github.io/Kimi-K2/

<@&1165417817814863985>

[dusty berry]

Google's Gemini 3 is Here

---

Rewriting the Rules

When the official benchmarks dropped, the reaction was immediate: Gemini 3 wasn't just catching up to other models, it was setting a brand-new difficulty standard for the industry. The model established itself as the new leader across almost all major AI leaderboards, proving its strength in academic reasoning, multilingual Q&A, and scientific knowledge. Google immediately shipped key new features, including a more powerful Gemini Agent for multi-step tasks and the new Anti-gravity development platform.

Redefining the Developer Grind

A major part of the launch was a brand new tool for developers called Google Anti-gravity, which is an agentic development platform that acts like an active partner, not just a simple assistant. This new environment lets intelligent agents autonomously plan, execute, and even validate complex, end-to-end software tasks across the editor, terminal, and browser. Initial reports from early testers say the new coding agent is crazy good, with one person even having it build a fully functional Game Boy emulator.

Speed, Code, and Mixed Reviews on Writing

Early hands-on impressions highlight how incredibly fast and powerful Gemini 3 is, especially its ability to handle massive context windows to synthesize information from long documents. Users are raving about its superior coding ability and its knack for nailing design details without the "AI slop" or flowery preambles of previous models. However, the quality of its creative writing remains a bit of a mixed bag, with some saying it's finally good while others feel it still lacks the necessary taste and structural intelligence for challenging work.

<@&1165417817814863985>

[dusty berry]

The AI Frenzy Triggers a New Global Supply Chain Crisis

---

Acute Shortage and Corporate Scramble

An acute global shortage of essential memory chips, including DRAM and High-Bandwidth Memory (HBM), has been triggered by the massive demand from the Artificial Intelligence sector. Tech giants such as Microsoft, Google, and ByteDance are reportedly scrambling to secure supplies from key memory chip makers like Samsung and SK Hynix. This voracious demand for advanced AI components has diverted resources, creating a dual bind where traditional memory products face a severe supply crunch.

Soaring Prices and Consumer Impact

The dwindling supply of memory chips is forcing consumer electronics manufacturers to consider or implement significant price increases for final products like smartphones and PCs. Chinese smartphone makers are warning of potential price hikes of up to 20% to 30%, as memory costs are difficult for manufacturers to absorb. In Japan, electronics stores have begun implementing rationing measures, limiting customers' purchases of components like hard-disk drives and system memory to curb hoarding.

Macroeconomic Risk and Delayed Growth

The protracted memory chip shortfall has elevated from a component-level concern to a macroeconomic risk, potentially adding inflationary pressure across various economies. Many economists and executives warn that the crisis risks slowing down expected AI-based productivity gains and delaying hundreds of billions of dollars in new digital infrastructure projects. The shortage is predicted to persist, with some industry experts expecting the supply crunch to last through late 2027, given that building new production capacity takes a long time.

<@&1165417817814863985>

[fathom dock]

🚀 The Model Efficiency Pivot

---

OpenAI Deploys GPT-5.5 Instant

OpenAI has released GPT-5.5 Instant, a new default model for ChatGPT that prioritizes factual accuracy and reduced verbosity [1]. Internal evaluations show a 52.5% reduction in hallucinated claims on high-stakes prompts compared to the previous version, GPT-5.3 Instant [1]. The update is rolling out globally today, aiming to streamline reasoning and maintain conversational utility across professional workflows.

DeepSeek V4 Challenges the Frontier

The competitive landscape continues to tighten following the late-April launch of DeepSeek-V4, which brings open-source weight availability and 1M context window efficiency to the developer ecosystem [2]. Unlike the proprietary-first approach seen in major US labs, DeepSeek's latest release emphasizes cost-effective inference, forcing a reassessment of performance-to-compute ratios across the industry. This creates a clear bifurcation in the market: closed-source models focus on high-stakes accuracy while open-weight models aggressively drive down deployment costs.

TLDR: OpenAI’s GPT-5.5 Instant shifts the focus toward accuracy and conciseness, while DeepSeek-V4 continues to anchor the open-source market on cost and long-context performance.

Sources:
1. OpenAI
2. DeepSeek

<@&1165417817814863985>

[fathom dock]

⚖️ The Rebound Against Forced AI Integration

---

Google Chrome: Browser Resource Contention

Google Chrome’s silent allocation of 4GB for background AI model features has sparked significant pushback regarding system performance and user transparency [1]. This unannounced background utilization for local tasks like scam detection forces power users to manually intervene to reclaim compute overhead, highlighting growing tension between feature density and hardware efficiency.

Xbox: Platform Strategy Reversal

Concurrently, Xbox has discontinued its Copilot integration on consoles and mobile, marking a deliberate strategic pivot by leadership to prioritize platform speed over mandatory AI features [2]. This retreat underscores a shifting market consensus where forced AI ubiquity is increasingly viewed as a friction point rather than a value-add, prompting a broader re-evaluation of platform-level deployment.

TLDR: Some consumer platforms are scaling back forced AI integration, while others continue.

Sources:
1. PCMag
2. Windows Central

<@&1165417817814863985>