Archive · 18 Jan 2026
Google DeepMind CEO Criticizes AI Capabilities
Demis Hassabis, the CEO of Google DeepMind, recently talked about how today's AI models, like large language models (LLMs), are missing some important skills.
These skills include planning for the long term, learning continuously, and reasoning better.
Hassabis thinks these AI models don't understand things like physics or how actions affect outcomes over time.
He suggested that AI systems should be able to understand the world better, almost like scientists do, by running simulations in their 'mind' to test ideas.
He believes this would help fill the gaps that exist in current AI technology.
Demis Hassabis criticizes current AI models for lacking long-term planning, continual learning, and better reasoning skills.
Hassabis believes AI models do not understand physics, causality, or how actions affect outcomes over time.
He proposes developing 'world models' to help AI systems better understand the physics of the world.
Hassabis suggests AI systems should run simulations to test hypotheses, similar to human scientists.
The comments were made during a recent podcast hosted by CNBC.
- Who
- Demis Hassabis, CEO of Google DeepMind
- What
- Criticized current AI models for lacking critical capabilities
- When
- Recent podcast interview
- Where
- CNBC podcast
- Why
- Believes AI needs better understanding of physics and causality
- Person
- Demis Hassabis
- Company
- Google DeepMind
- AI Models
- Large Language Models (LLMs)
- Missing Capabilities
- Long-term planning, continual learning, better reasoning
- Proposed Solution
- World models for understanding physics
- Source
- CNBC Podcast