Big Shots Weekly · Ep. 18: Anthropic's Public Fund, Sam Altman's Navier-Stokes, and Microsoft's On-Device Intelligence

It's Friday roundup time again, and the big shots were busy this week. Anthropic is putting shares into a public fund, Sam Altman is showing off math problems, and Microsoft announced a major Windows update. And if you look at these together, it's really interesting. One is about how money gets spent, one is about where the capability limits are, and one is about whether AI runs in the cloud or on your own computer. Let's start with the biggest one. Anthropic's official Twitter announced they're putting some shares into a new Anthropic public fund — a wholly owned entity dedicated to public interest work. It starts with 5% of shares. Then for every 50% increase in company valuation, they can add up to another 2%, decided by the board. Altogether, about 20% of shares are committed to the public interest. 20% — that's a big number. And they said the co-founders have already pledged to donate 80% of their personal equity, and about half the employees are donating too. Right, and here's the key point. This isn't like normal corporate charity — they're locking the shares themselves into an entity. The more valuable the company gets, the more money the fund has. It's a structural commitment. So why now? Anthropic's been moving fast lately — releasing models, running scholarships. Think about it — the stronger the models get, the more scrutiny AI companies face. By opening up their interest structure proactively, they're sending a signal to regulators and the public: we're not just here to make money. Same day, they opened applications for two new scholarships — one in STEM, one in conceptual reasoning, covering philosophy, cognitive science, economics, policy, and law. I really like this. The conceptual reasoning group is meant to help Claude think more clearly about the hardest humanities and social science questions. See — donating money on one side, hiring people on the other, both centered on the word 'trustworthy.' Okay, next up, Sam Altman. He tweeted a Navier-Stokes milestone T-shirt, with the last line of the chain of thought printed on the back. The Navier-Stokes equations — one of the seven Millennium Prize Problems. Printing the last line of the AI's chain of thought on a T-shirt is so Sam — showy, and a little romantic. Wait, what does that last line of the chain of thought actually mean? Did the model really make progress on the problem? The way he posted it, it's less a declaration of solving it and more a showcase of the reasoning process itself. The point isn't the answer — it's that AI can walk through a derivation at that level, step by step. And that connects to another OpenAI item — they donated five million dollars to the Simons Foundation, specifically to support mathematicians in understanding and verifying AI outputs. Right, and putting those two together makes sense. On one side AI can produce results, on the other they're paying human mathematicians to verify them. They know outputs alone aren't enough — someone has to check. So this is basically an official admission that AI's math output needs humans to digest and consolidate it? I think so. And honestly, that's pretty candid. The capability is there, but trustworthiness needs humans to close the loop. That five million isn't buying compute — it's buying academic trust. Then there's another OpenAI one — GPT-Rosalind found new leads in drug discovery, involving a gene called DCTN1 linked to Perry syndrome. This case is very typical. It proposed a mechanistic hypothesis: cholesterol might not be reaching where neurons need it. Then thousands of agents ran in parallel and in thirty-two minutes connected gene expression results from millions of brain cells with the literature. Thirty-two minutes? I have a hard time believing that. How long would a human team take for literature and data correlation at that scale? That's the power of parallel agents. But note — it gave a testable mechanism, not a conclusion. The real value is pointing scientists in a direction and saving months of exploration. Okay, next, Microsoft. Satya Nadella tweeted that Windows is entering a new chapter — bringing unlimited intelligence to every desk and every home. This one's the most packed. He announced MAI-Code-1.1 Flash — 137 billion parameters, 256K context, and the key thing is it's optimized to run directly on your PC. Run on a PC? So no cloud token costs? He said Code in Copilot can build any software on your desktop — the PC becomes an infinite software factory. Right, that's the Hybrid Intelligence idea. Sensitive work happens locally, and Copilot can act directly on the PC. GitHub Copilot hands tasks to local models, cutting costs a lot without losing quality. What about security? An agent moving things around on your computer — as a product manager, my first reaction is risk. He specifically mentioned it — Windows is integrated with Agent 365, plus a local sandbox called MXC where agents execute. He wants Windows to become your secure agent box. Stringing these together, I get the feeling the big companies are telling the same story this week: AI isn't just a chat box in the cloud anymore — it's coming to your device, with an accountability framework attached. Exactly. Anthropic uses share structure to show responsibility, OpenAI pays people to verify, Microsoft uses a local sandbox to manage agents. Three different angles, same direction: the stronger the capability, the more you have to talk about how to be trusted. So for ordinary users and developers, what's the most practical impact? Short term, running models locally will drastically change the cost structure of agent products — the old burn-money-on-tokens approach will get disrupted. Long term, whoever builds a solid trust framework will get into enterprises. So to sum up this week's signals: public fund, math verification, on-device intelligence — all three lines are really answering the same question: now that AI has grown up, how do we use it with confidence? And notice a trend — the contrast is obvious. Some are adding to public responsibility, some are pushing the capability ceiling, and some are laying infrastructure. I bet there'll be more next week — we'll keep watching.