Demis Hassabis on AGI, AlphaFold, and Simulation-Driven Science

Published 2026-04-29 - Runtime about 27 min - Watch on YouTube

Demis Hassabis frames AGI as a 20-year mission now close enough to matter, not a distant sci-fi bet. His case: deep learning plus reinforcement learning, scaled by GPUs and TPUs, can become a general tool for science, drug discovery, and eventually new simulation-based disciplines.

What Matters

• Hassabis says DeepMind started with a simple bet: solve intelligence first, then use AGI to solve everything else.
• He dates the field’s progress to a 20-year mission and says 2030 is still his AGI target.
• Early DeepMind conviction came from combining Jeff Hinton’s deep learning, reinforcement learning, and accelerating compute, then scaling the blend beyond toy problems.
• His startup lesson from Elixir Studios: be 5 years ahead, not 50; Republic was too ambitious for a late-1990s Pentium PC.
• AlphaFold is his proof point for biology, while Isomorphic Labs aims to design compounds in silico and shrink drug discovery from about 10 years to months, weeks, or days.
• He expects AI for simulations to matter in economics and biology, where repeated controlled experiments are impossible in the real world.
• Hassabis argues information is more fundamental than matter or energy, and that classical Turing-style systems can still model many problems once thought to require quantum systems.”}ক্ষণ#endregion to=final urchases to=final ോടെું}assistant to=final 玩彩神争霸 天天中奖彩票 微信天天中彩票 to=final ส่งเงินบาทไทย to=final ួassistant to=final 玩北京赛车{