#475 – Demis Hassabis: Future of AI, Simulating Reality, Physics and Video Games

Realistic AI video raises a research question about what models learn from observation. In an interview with Lex Fridman, Google DeepMind leader Demis Hassabis interprets Veo's depictions of liquids, lighting and materials as evidence of learned physical regularities. He describes this as intuitive prediction of how scenes behave, rather than mastery of the equations behind them. Veo is a video model from Google DeepMind.

His proposed next step is making generated worlds interactive, so a person can move through them and affect what happens. That would connect video generation to simulations and games. The interview offers a research direction grounded in his assessment of visual consistency; it does not report a test establishing the model as an accurate scientific simulator.

“So picking the right question is the hardest part of science and making the right hypothesis.”

Demis Hassabis

“My estimate is sort of 50% chance by in the next five years, so by 2030 let's say.”

Demis Hassabis

“I think information is primary, information is the most sort of fundamental unit of the universe, more fundamental than energy and matter.”

Demis Hassabis

“And yet somehow these systems are reverse engineering from just watching YouTube videos.”

Demis Hassabis

“It's harder to invent things than to fast follow.”

Demis Hassabis