The Artificial Human Are Large Language Models a Dead End?
11 snips
Feb 25, 2026 
Michael Woolridge, Oxford AI professor critiquing LLMs as predictive hacks. Jeff Hawkins, neuroscientist behind Thousand Brains Project building brain-inspired, sensory-motor AI. They debate LLM limits, embodiment, world models, reference frames and sensory-driven learning. Short, spirited conversations explore alternatives to text-only approaches and how real-world understanding might be achieved.
AI Snips
Chapters
Transcript
Episode notes
LLMs Are Powerful Yet Narrow
- Large language models (LLMs) are impressive and accessible but focus narrowly on language prediction rather than broad embodied intelligence.
- Michael Woolridge notes LLMs do token prediction from text corpora and lack the multimodal, action-oriented capacities humans evolved for.
Fluency Doesn't Equal Practical Competence
- Despite fluent outputs across many topics, LLMs fail at real-world embodied tasks like navigating a kitchen or loading a dishwasher.
- Woolridge contrasts fluent text about the Roman Empire with absence of robotic capability to show multifaceted human intelligence.
Current AI Research Feels Like Alchemy
- The current AI boom resembles alchemy: powerful tools we don't fully understand but which may still lead to useful science.
- Woolridge warns much contemporary AI work is 'prodding' models to observe emergent behaviours without deep explanation.