Everything in AI/tech this week in 5 minutes

Happy Sunday. Here's what actually mattered in tech this week, in about five minutes, then a bit on what's been on my mind. Let's get into it.

This week in tech + AI

OpenAI claims one of its internal AI models produced a proof for the Navier-Stokes problem, one of the seven Millennium Prize Problems in math. On September 8 the company said an internal model generated a proof, formalized in the Lean proof assistant, showing that the Navier-Stokes equations can develop a singularity in finite time. According to OpenAI it ran roughly 10,000 AI agents in parallel and reached the result in about 88 hours. The work is now being reviewed by the math community, and some mathematicians have publicly raised questions about how much of the approach overlaps with earlier unpublished work, so it is a potentially historic moment for AI in math that is still being checked. Read more

An Anthropic researcher resigned with a public warning about AI safety. Jacob Coxon, who says he spent three years doing research at Anthropic and OpenAI, announced his departure on X and wrote that the leading labs are racing each other on capability faster than they are addressing safety. A couple of current Anthropic employees publicly responded to his posts, which is part of why it spread. It is a notable moment in the ongoing public debate about how fast frontier AI should move. Read more

Mistral raised 3 billion euros, the largest equity round in European tech history. The French AI lab closed a Series D at a post-money valuation above 21 billion euros, led by Samsung, three years after it launched. A good reminder that the frontier model race is not only a US story. Read more

On my mind

Don’t get stuck in local maxima. I was recently talking to an investor from a top European VC to get some guidance on my path forward, and one thing he said really stuck with me. He told me to avoid getting stuck in a “local maximum”, a term you hear a lot in optimization problems. Basically he was warning me against focusing too much on exploiting my current situation (my network, my ideas, my topics) without exploring new things enough. Maybe there is an even better idea, area of research or network hiding somewhere I haven’t looked yet, and the only way to find out is to stay open-minded and curious. I really liked this analogy borrowed from how ML models are trained, and I strongly agree with his point. So to avoid getting stuck in a local maximum, you have to stay curious: talk to new people, go to new events, learn about new topics and so on.

More from me

🎬 My latest content:

Hit 300k and took a minute to reflect on everything that has happened since I started posting. Have a look
The four tools I actually use every day to stay productive as a student, creator and founder. Have a look

👥 Community Section:

A reader put me onto freeCodeCamp for learning to code with its free certifications, and since I use it too I'm happy to pass it on. A great way to start from zero. freecodecamp.org
A friend is hiring an experienced reinforcement learning engineer, San Francisco or remote, paying $10,000-$15,000 a month. If you have genuinely strong RL skills you can actually prove with real work, not just standard course projects, email me your resume and links ([email protected]) and I'll forward you. Please refrain from emailing if you have little experience in RL and don’t have tangible proof (publications, open-source contributions, internships in RL)

Got something for this section, a resource, a job opening, a co-founder search? Email me at [email protected] with the subject "Newsletter Community Section Idea".

🔨 Tool of the week: Hubspots repot about the Future of AI Marketing