Welcome, humans.

Okay, so apparently someone got tired of staring at Claude Code while it thinks and built a literal waiting room for it.

It’s basically Omegle for Claude users: when you’re waiting on Claude, it matches you into a voice/video chat with another random person who is also waiting on Claude, so you can complain, compare what you’re building, or maybe accidentally make a friend.

And in the most meta finding of all, the creator vibe-coded the whole thing with Fable 5.1 over Labor Day weekend.

So yes, we have officially reinvented Chatroulette for people waiting on their AI to finish coding. This is peak agentic internet people!!

Here’s what happened in AI today:

• 🙀 OpenAI says 10,000 agents solved Navier-Stokes.

• 📰 DeepMind mapped all 9 billion possible DNA mutations.

• 📰 Ant released an open finance model for analysts.

• 🍪 ChatGPT Images 2.5 rolled out across all tiers.

• 🎓 Three new ways to use ChatGPT on desktop.

…and a whole lot more that you can read about here.

🙀 OpenAI says 10,000 AI agents solved a $1M math problem

Okay, so OpenAI says an unreleased model significantly more capable than GPT-6 Astra helped solve one of mathematics’ seven famous Millennium Prize Problems.

The problem is called Navier-Stokes, and it asks whether equations used to model fluids can eventually “blow up,” producing an infinite velocity even when everything starts smoothly.

Mathematicians have been stuck on it for decades. OpenAI says its system found a proof in roughly 88 hours.

Here’s what happened:

• OpenAI unleashed roughly 10,000 agents at once, giving them tools, code execution, and the ability to share useful discoveries.

• The effort generated 2.7M agent messages and roughly 130B output tokens.

• The agents found a proposed solution showing that a fluid can develop a singularity in finite time.

• GPT-6 Astra then helped formalize and verify the proof in Lean, software that checks mathematical proofs step by step.

And yes, there’s drama. NYU mathematician Tristan Buckmaster and Anthropic researcher Levent Alpöge were already working on related fluid-dynamics problems.

OpenAI says it heard rumors of breakthroughs, then pointed thousands of agents at the remaining Millennium Problems. It denies accessing the researchers’ work and says its proof was independently produced.

If you want the academic knife fight, TechCrunch and WIRED unpack the allegations, while Scientific American digs into what mathematicians think of the proof itself. Another Scientific American piece compares the moment to Deep Blue beating Garry Kasparov.

But the part I can’t stop thinking about is the 10,000 agents. One model did not sit there and have a Newton-under-the-apple-tree moment. OpenAI bought an enormous amount of parallel thinking.

Our take: Start judging frontier AI by what a system can accomplish with enough agents, tools, time, and compute, not just what one chatbot answers in one conversation. Now again, in President JFK accent: Don’t ask what your agent can do for your country… ask what a country of agents can do for you!

OpenAI already says the model behind this is better than Astra. Pretty soon (pending regulation notwithstanding) we get to find out what happens when everyone else gets access to it. Will 1000 unsolved math problem shaped flowers bloom? Or will we all just make more blender demos?

FROM OUR PARTNERS

Build Compliance into your AI Stack with Vanta

What could your GRC team do with 12 hours back each week? Vanta connects controls, evidence, and risk data in one place, using tools you already use like Claude, Cursor, and Codex.

Register for this live demo to see how to:

• Unify controls, evidence, and risk data in the Trust Graph

• Connect Vanta into Codex, Claude, and Cursor

• Create custom agents and skills inside your existing tools

→ Save your seat

🎓 AI Skill of the Day: 3 ways to use ChatGPT in a totally new way

The new GPT-6 Astra showed off some impressive capabilities, especially around voice and computer use. If you want to be like Corey over Labor Day weekend and vibe code on your phone while you’re out back barbecuing, follow these tips below!

First and foremost, if you haven’t tried the new ChatGPT desktop app yet, download it here.

Want to direct ChatGPT like the GPT-6 launch vid? Use Voice + computer use like so:

1. Open ChatGPT desktop, choose Work or Codex in the top left, then go to → Plugins → Computer Use. Install/enable it, turn on the Computer Use server + skill toggles, and approve the permissions it asks for.

2. Start a task, turn on Voice, and tell ChatGPT to use Computer Use.

3. Then talk it through the job while it controls your computer: tell it to open apps, click, type, make changes, check progress, change direction, or explain what it’s doing. I do this like all the time.

Next up: want to direct it to do things, but on your phone? You need Remote control: 

1. Start a Codex task on your computer, then enable Remote Control in desktop settings and toggle “Allow Connections” to on.  

2. Open the ChatGPT mobile app → Remote and connect to that desktop session with a QR Code. Pro tip: you can do this with 2+ computers!

3. Now, your computer does the actual work while you steer it from your phone. Ever have a boss? This is how your boss feels making you do things.

Uh oh! You just got vibe-code pilled and hit 0% remaining on your usage for the week! Been there too many times to count, friend. We got you

Do you have ChatGPT account from prior to GPT-6 launch? Use your Banked resets!

1. Go to Settings → Usage (sometimes shown as Usage & limits).

2. If you have a banked reset (or three), select the available reset, check when it expires, and click it twice to confirm using it.

3. A full reset refreshes your weekly Codex windows, then moves your weekly reset date seven days from now. Use sparingly… and follow Tibo (aka Mr Reset himself).

Have a specific skill you want to learn? Request it here.

FROM OUR PARTNERS

Your AI agent won't flag an outdated refund policy. But it’ll state it confidently to thousands of customers before anyone notices. Guru's new ebook, The Knowledge Accuracy Gap, explains why hiring more reviewers won’t work, and the two-layer architecture that verifies knowledge before any agent reaches it. 

Close your accuracy gap

📰 Around the Horn

• Google DeepMind released AlphaGenome Atlas, predicting the molecular effects of all 9 billion possible single-letter changes in the human genome.

• Ant Group open-sourced Ling-3.0-flash-Fin, a finance model built for research, filings, valuation spreadsheets, and cited reports with only 5.1B parameters active per token.

• Cognition raised more than $2B at a $48B valuation to keep scaling Devin and its software-engineering platform.

• Google Cloud and Accenture created a Gemini Enterprise unit that plans to put 1,000 forward-deployed AI engineers directly inside customer organizations.

• Qualcomm and Amazon struck a multi-generation deal to build custom AI inference chips and high-speed connectivity for AWS data centers.

• TSMC and Samsung committed to ASML’s newest High NA EUV chipmaking technology as the industry pushes toward even smaller, denser AI chips.

• Google recently helped map all 166,000 neurons in a male fruit fly, and developers immediately turned the wiring into software that classifies human emotion, simulates pain, and powers NPCs in a browser game.

🍪 Treats to Try

*Asterisk = from our partners (only the first one!). Advertise to 700K+ readers here!

1. AvePoint's AI Virtual Summit (Sept 10) makes the case that you don't need new rules for agentic AI, you need to apply the ones that already work; get an auditability framework you can use. Save your free seat

2. Meta launched Muse, a U.S.-only personal agent that runs long jobs in its own cloud VM, with a free tier plus $20 and $100 monthly plans.

3. ChatGPT Images 2.5 gives you more control when editing existing images while doing a better job preserving faces, composition, and visual identity.

4. LLM Attention Visualization lets you actually see which words a language model pays attention to as information moves through its layers.

5. I Have ADHD is a coding-agent skill that forces Claude and other agents to stop burying the answer under giant walls of explanation.

6. Copperhead is basically Cursor for circuit boards: describe the hardware you want, and its open-source agent designs, documents, and validates the PCB.

7. Switch brings Claude Code and other AI agents into Slack, Teams, Discord, or Telegram as named members of the conversation.

📖 Midweek Wisdom

• Cybersecurity officials argued that boring security basics still stop most attacks: identity management, perimeter monitoring, cyber hygiene, and strong multi-factor authentication.

• IEEE Spectrum asks what happens to junior engineers when AI writes more code and humans increasingly get promoted from builder to reviewer.

• Wharton digs into a weird AI productivity problem: companies can generate software faster, but human review and deployment bottlenecks keep much of it from reaching users.

• The Atlantic argues medicine should start seriously testing increasingly autonomous AI doctors instead of treating the idea as science fiction.

• The Wall Street Journal explores how AI can combine tiny public clues across the internet to work backward from a supposedly anonymous account to the person behind it.

• Dwarkesh Patel and Jerry Han estimate that from 2019 to 2025, better training data drove about 12× pretraining efficiency gains, versus about 3.7× from model architecture and training-recipe improvements.

• Chamath Palihapitiya broke down Anthropic’s roughly $80B in neocloud commitments over five days, including $45B with Nscale and $35B with Lambda, and the financing risk behind the boom.

BTW: I went down a rabbit hole on Kyle Jeong’s excellent breakdown of how Astra controls a computer, and the basic loop is surprisingly understandable:

1. Observe: Astra reads screenshots and the computer’s accessibility tree, a text map originally designed to help people with disabilities navigate interfaces.

2. Act: it writes code telling the browser or computer what to click, type, or open.

3. Verify: it looks at the screen again to make sure the action actually worked.

Then it repeats. And if you didn’t know, OpenAI has an API for this so other apps (like one you build, for example) can make this feature available. 💫 the more you know!

A Cat’s Commentary

Btw: We just launched a robotics newsletter! Sign up for it here.