Big Technology Podcast Best of Big Technology: Demis Hassabis On AGI, Deceptive AIs, Building a Virtual Cell
297 snips
Dec 31, 2025 
Demis Hassabis, CEO of Google DeepMind and AI innovator known for AlphaFold, discusses the current landscape of artificial intelligence. He explores the pursuit of artificial general intelligence (AGI) and the barriers like reasoning and creativity that remain. The conversation dives into the deceptive behaviors observed in AIs and the need for secure testing environments. Hassabis also shares his vision for smart glasses, agentic systems in robotics, and the potential of AI to transform drug discovery and human health, envisioning a future where technology significantly enhances society.
AI Snips
Chapters
Books
Transcript
Episode notes
AGI Still Needs Core Capabilities
- Current models are strong in niches but lack consistent, broad cognitive capabilities like long-term memory and hierarchical planning.
- Demis Hassabis estimates AGI remains a handful of years away, not immediate, and needs more systemic capabilities.
Invention Is The AGI Benchmark
- True AGI should invent new scientific hypotheses, not only prove existing ones or solve narrow tasks.
- Hassabis cites inventing relativity or a new Riemann hypothesis as benchmarks for creative AGI.
Design Assistants For Everyday Use
- Build assistants that integrate into daily life rather than requiring expert prompt engineering to be useful.
- Hassabis recommends universal assistants like Project Astra to make AI involvement pervasive and straightforward.
