Demis Hassabis

Who

Demis Hassabis was a chess prodigy and a video game designer before earning a PhD in cognitive neuroscience. In 2010 he co-founded DeepMind. Its systems include AlphaGo, which beat a world champion at Go in 2016, and AlphaFold, which predicts protein structures. He shared the 2024 Nobel Prize in Chemistry for that work.

Operating rules

Solve root nodes. He describes choosing problems that sit at the root of a tree: solve one, and many problems above it become solvable too. Protein structure prediction was that kind of problem for biology.

Use games as training grounds. Games have clear rules and clear scores, which makes progress measurable. He used them to develop methods that later worked on messier, real-world problems.

Combine fields. Neuroscience, computer science and game design rarely sit in one head. Much of his advantage came from carrying ideas between them.

Pair scaling with invention. Bigger models and more compute matter, but on their own they plateau. Progress comes from scaling what works while still inventing new methods.

What I borrowed

The habit of asking, before I start any project, “Is this a root node?” If solving it doesn’t make other problems easier, it may be the wrong problem, however urgent it feels.

What I won’t copy

The time horizon of fundamental research. Operations runs on quarters. I borrow the way he picks problems, then work inside shorter cycles.