Language models can't spark scientific revolutions, but world models might
Google DeepMind argues LLMs hit a cognitive ceiling — but world models might break through it.
Why it matters
A frontier lab researcher challenges the current path to AGI, claiming language models lack the mechanism for true scientific innovation. This shapes how practitioners and labs think about which model architecture to invest in next.
The key facts
8 to knowTom Zahavy (Google DeepMind) published position paper 'LLMs can't jump'
Claims LLMs lack cognitive mechanism for genuine novelty creation
Suggests world models as potential alternative architecture
Published July 30, 2026
Tom Zahavy (Google DeepMind) position paper: 'LLMs can't jump'
Argument: LLMs lack cognitive mechanism for generating truly novel scientific insights
Implication: World models positioned as the next frontier for scientific AI capability
Published Jul 30, 2026
Go to the source
The Decoderthe-decoder.com
Publisher excerpt: Can language models spark a scientific revolution? In a position paper titled "LLMs can't jump," Google Deepmind's Tom Zahavy argues they can't. They're missing the cognitive mechanism needed to create something truly new.