The Man Who Turned Down $1.3 Million to Stay in School — And Changed Medicine Forever

What does it actually take to solve a scientific puzzle that stumped researchers for fifty years — and why are the same AI tools that cracked it still getting basic financial math catastrophically wrong?

In 2024, Demis Hassabis won the Nobel Prize in Chemistry for something that sounds almost too ambitious to be real: teaching artificial intelligence to predict the three-dimensional shape of any protein, a problem that had resisted traditional science for half a century. The breakthrough, called AlphaFold, has already mapped over 200 million proteins, a feat with profound implications for how we understand and treat disease.

In this episode, Larry Kotlikoff sits down with Sebastian Mallaby, the New York Times bestselling author and veteran journalist behind The Infinity Machine, his deeply reported new biography of Hassabis and the rise of DeepMind. Mallaby walks through the improbable arc of Hassabis's career - from chess prodigy to video game programmer to the architect of some of the most consequential AI breakthroughs of the last decade. Then turns to a sharper, more urgent question: what can AI actually be trusted to get right, and where does it quietly, confidently fail?

What You'll Learn:

• [00:32:01] The $1.3 million check Demis Hassabis turned down — and why he chose to study computer science instead of cashing in on his teenage video game fortune
• [00:33:00] Why AI was considered a dead field as recently as 2012 — and how Hassabis convinced Peter Thiel to fund a company built on a technology that, at the time, couldn't do much of anything
• [00:34:05] What "agentic AI" actually means — and how DeepMind's early systems learned to master Atari games through pure trial and error, years before "agentic" became a buzzword
• [00:35:08] Reinforcement learning vs. deep learning, explained simply — the difference between an AI that learns from reading and one that learns from doing
• [00:43:56] How neural networks actually work — a plain-language breakdown of "weights," layers, and how a system tweaks itself toward a correct answer through millions of small adjustments
• [00:41:00] Why AI is fundamentally inductive, not deductive — and why that distinction explains both its remarkable strengths and its most dangerous blind spots
• [01:04:57] How AlphaFold's protein-folding breakthrough connects directly to disease research — including new frontiers in understanding conditions like Parkinson's
• [01:05:59] Inside Isomorphic Labs — Hassabis's newest venture, aimed at using AI to dramatically accelerate (not replace) the drug discovery pipeline
• [01:16:00] Why AI tools are getting deterministic financial questions wrong — Larry's own real-world tests show major AI models disagreeing wildly on life insurance needs, optimal savings, and Roth conversion strategy, even though these problems have one mathematically correct answer
• [01:12:43] What this means for the future of work — a candid exchange on AI adoption timelines, productivity's "jagged edge," and whether human labor markets have time to adjust

Featured Guest:

Sebastian Mallaby is an award-winning journalist and author of six books, including the New York Times bestseller The Infinity Machine: How an AI Pioneer Reshaped the Future, as well as The Power Law: Venture Capital and the Making of the New Future and More Money Than God. A former correspondent for The Economist and longtime member of the Washington Post editorial board. He co-hosts the podcast The Spillover with Rebecca Patterson of the Council on Foreign Relations.