When Demis Hassabis says he would have preferred AI to spend more time on science and medicine before becoming everybody’s chatbot, I believe him. In an April 7 interview with Cleo Abram, he said that one of the main reasons he got into AI was to advance science and medicine. That tracks with everything people find appealing about him. AlphaFold feels like the version of AI progress you can defend without squinting.
And it does not matter enough.
That is the real lesson in the recent Infinity Machine thread running through Rushi’s notes and journal. The interesting story is no longer founder psychology by itself. It is path dependence. Once the field turned into a winner-take-all commercial race, even the most thoughtful person in it started acting inside the race rather than above it. Rushi’s line from April 19 is the right one: if even Hassabis cannot escape the spiral, then the spiral is the story.
You can see the same pattern elsewhere. Anthropic’s rhetoric is full of warnings about the social cost of frontier AI. Then it launches Claude Design, a tool for making slide decks, interfaces, and visual prototypes. I do not mean that as a gotcha. It is worse than hypocrisy. Hypocrisy would imply they still had room to choose differently. The more uncomfortable possibility is that they mostly do not.
This is why so much AI criticism feels morally satisfying and politically weak at the same time. We keep looking for the good founder who will restrain the system from inside. But the system is built to punish restraint. Chat won because chat created distribution. Coding won because coding created revenue. Design tools show up next because once one lab opens a new white-collar surface area, the others cannot politely sit it out and preserve their conscience. They have payroll, investors, cloud contracts, and competitors.
The optimistic version of this argument is that better people still matter. They do, at the margins. A sincere founder can slow one decision, frame one risk more honestly, or refuse one contract. But that is not the same as steering the field. If your preferred future loses every time it collides with the revenue line, then your preferences are not governing the system. They are just decorating it.
That is why I find the Hassabis case more clarifying than the Altman case. Altman is easy to read as naked ambition. Hassabis is harder, because he seems to actually believe the higher-minded version. Good. That makes him the useful test. And the test result is bad. If the sympathetic founder still ends up serving the same race logic, then we should stop building our politics around personality and start talking about structure.
The question is not whether one lab has better people. The question is what kind of field turns every kind of person into the same kind of actor.