Big day in AI land, and I mean big. A model went and disproved a math problem nobody could touch for eighty years, one of the most respected names in the field jumped ship to a competitor, and half the industry is suddenly racing to ring the bell on Wall Street. Let me break it down.
AI Disproved an 80-Year Math Problem on Its Own
OpenAI says one of its internal models took an old geometry conjecture, the kind that has sat unsolved since before my granddad was born, and disproved it. Not with a hint from a researcher. On its own.
Tim Gowers, a Fields medalist, called it a milestone in AI mathematics. When a guy with that kind of hardware on his shelf says milestone, you listen.
Why does this matter? Because there is a world of difference between a chatbot that helps your kid with algebra and a system that finds something every human mathematician missed for eight decades. The first is a tool. The second is starting to look like a colleague.
My take: I have been telling folks the real story in AI is not the chat window, it is what these things can do when you point them at a hard, narrow problem. This is the proof. We are going to see a lot more of it, and a lot of smart people are about to feel a little less special. That is fine. Better tools have always made better work.
Karpathy Packed Up and Joined Anthropic
Andrej Karpathy, an OpenAI co-founder, is heading to Anthropic to work on the pre-training team. The plan is to use Claude to speed up pretraining research, which is a bit like using a forge to build a better forge.
This one matters because talent in this business does not move quietly. When a founder leaves the company he helped start and walks across the street, it tells you where he thinks the interesting work is happening.
My take: Folks will read tea leaves on this all week. I keep it simple. Smart people go where the problems are good and the leash is long. Anthropic landing Karpathy is a real feather in the cap, and OpenAI losing him is not nothing. Both can be true.
Everybody Is Suddenly Lining Up to Go Public
The money news is loud. OpenAI filed a confidential S-1 aiming for a September listing somewhere between $852B and $1T. Anthropic is targeting October at a $900B valuation and says it is on track for its first ever quarterly operating profit, with around $10.9B in Q2 revenue. NVIDIA, the company selling the shovels, just posted a record quarter with $81.62B in revenue.
Why it matters? Going public means opening the books. For an industry that has run on promises and private rounds, that is a big shift. Real numbers, real scrutiny, real accountability.
My take: I am glad to see it. Hype is cheap and an S-1 is not. When these companies have to show actual revenue and actual costs to actual investors, we all get a clearer picture of what is real and what is a slide deck. The profit line at Anthropic is the part I am watching. Talk is one thing. Black ink is another.
That is the roundup for today. Three stories, one theme: AI quit being a science project this year and started acting like a business with teeth. Catch you tomorrow.