How to Build the Future: Demis Hassabis

Why It Matters

Hassabis’ roadmap signals that breakthroughs in continual learning and efficient model distillation will determine when AGI‑level agents become mainstream, reshaping product strategy, infrastructure investment, and competitive advantage across industries.

Key Takeaways

• Continual learning, long‑term reasoning, and memory remain unsolved for AGI.
• DeepMind’s Gemini builds on reinforcement‑learning agents from AlphaGo era.
• Model distillation enables small, fast models without sacrificing most performance.
• Scaling may close gaps, yet one or two ideas remain needed.
• Efficient edge models could power privacy‑focused AI and robotics.

Summary

Demis Hassabis, DeepMind CEO, outlined the current roadmap toward artificial general intelligence, emphasizing that while large‑scale pre‑training, RL‑HF and chain‑of‑thought have propelled capabilities, core ingredients such as continual learning, long‑term reasoning and robust memory systems are still missing. He positioned Gemini as the next iteration that fuses the agent‑centric reinforcement‑learning heritage of AlphaGo, AlphaZero and AlphaStar with modern foundation‑model techniques, arguing that the same principles of goal‑directed search and Monte‑Carlo planning remain central.

Hassabis highlighted concrete progress: AlphaGo’s historic victory, AlphaFold’s protein‑structure breakthrough (Nobel‑winning), and the early use of experience replay inspired by hippocampal consolidation. He noted that DeepMind now excels at model distillation, turning frontier‑scale models into “flash” versions that retain ~95% performance at a fraction of the cost, a capability essential for serving billions of Google products and for edge deployment.

The interview also featured candid observations about current limits. Large context windows, though massive compared with human working memory, still operate as brute‑force storage, making retrieval costly. Chain‑of‑thought reasoning can still loop or make elementary errors, suggesting that more sophisticated introspection or monitoring of thought traces is required. Hassabis estimates a 50/50 chance that one or two novel ideas—not just scaling—will be needed to bridge these gaps.

If DeepMind’s predictions hold, the next decade could see AGI‑level agents integrated into everyday services, from real‑time personal assistants to privacy‑preserving robotics. The push for efficient, locally‑run models may reshape hardware investments and regulatory discussions, while the timeline Hassabis cites—around 2030—forces enterprises to plan for rapid, disruptive AI adoption.

Original Description

Demis Hassabis has had one of the most extraordinary careers in tech. He started as a chess prodigy and video game designer at 17 before getting a PhD in neuroscience and going on to found DeepMind. His lab cracked Go, solved protein structure prediction with AlphaFold, and then gave it away free to every scientist on earth. That work won him the 2024 Nobel Prize in Chemistry. Today he leads Google DeepMind, pushing toward the same goal he set as a teenager: AGI.

On this special live episode of How to Build the Future, he sat down with YC's Garry Tan to talk about what still needs to happen to get us to AGI, his advice for founders on how to stay ahead of the curve and what the next big scientific breakthroughs might be.

Chapters:

00:00 — Intro

00:46 — Demis Hassabis: From Chess Prodigy to DeepMind

01:48 — What’s Missing Before We Get To AGI?

03:36 — Why Memory Is Still Unsolved

06:14 — How AlphaGo Shaped Gemini

08:06 — Why Smaller Models Are Getting So Powerful

10:46 — The 1000x Engineer

12:40 — Continual Learning and the Future of Agents

13:32 — Why AI Still Fails at Basic Reasoning

15:33 — Are Agents Overhyped or Just Getting Started?

18:31 — Can AI Become Truly Creative?

20:26 — Open Models, Gemma, and Local AI

22:26 — Why Gemini Was Built Multimodal

24:08 — What Happens When Inference Gets Cheap?

25:24 — From AlphaFold to the Virtual Cells

28:24 — AI as the Ultimate Tool for Science

30:43 — Advice for Founders

33:30 — The AlphaFold Breakthrough Pattern

35:20 — Can AI Make Real Scientific Discoveries?

37:59 — What to Build Before AGI Arrives

Apply to Y Combinator: https://www.ycombinator.com/apply

Work at a startup: https://www.ycombinator.com/jobs