DeepMind Researcher: LLMs Can't Spark Scientific Revolutions, But World Models Might

The Decoder · rss · 2026-07-30

In a position paper titled "LLMs can't jump," Google DeepMind researcher Tom Zahavy explores the limitations of Large Language Models in driving scientific breakthroughs.

He argues that LLMs lack the specific cognitive mechanisms required to create something genuinely new, meaning they cannot independently spark scientific revolutions. However, the article suggests that world models could fill this gap and serve as the key to unlocking fundamental innovations in the future.

Original post →

More from AGI Musings

AGI Musings channel →