Y Combinator Startup Podcast How to Build the Future: Demis Hassabis
646 snips
Apr 29, 2026 
Demis Hassabis, DeepMind co-founder and Nobel-winning AI researcher, dives into what still stands between today’s models and AGI. He explores memory, reasoning, agents, and why smaller local models are getting so powerful. The conversation also touches on AlphaGo’s influence on Gemini, virtual cells, AI for science, and what founders should build before AGI arrives.
AI Snips
Chapters
Transcript
Episode notes
Move 37 Changed How Demis Viewed Creativity
- Demis Hassabis uses AlphaGo’s Move 37 as his benchmark for machine creativity, but says true creativity would mean inventing an entire game like Go.
- He started AlphaFold right after returning from Seoul, because Move 37 proved systems could generate surprising, valuable ideas.
Why DeepMind Opens Its Edge Models
- Demis Hassabis argues open models matter strategically because edge models on phones, glasses, and robots are exposed anyway once deployed.
- He also wants strong Western open-source stacks, pointing to Gemma’s rapid adoption and competitiveness against Chinese open models.
Why Gemini Started Multimodal
- Gemini was built multimodal from the start because Demis Hassabis thinks real assistants and robots must understand physical context, not just text.
- He ties that choice to robotics, world models, Waymo, and devices like phones or glasses that operate in the real world.