DeepMind researcher says his “LLMs Can’t Jump” paper is about Einstein-style leaps, not AI dead ends
TZahavy · x · 2026-07-29
What the paper is and isn’t
The author says “LLMs Can’t Jump” is a personal position paper, not DeepMind’s official stance and not a claim that LLMs can never make scientific discoveries. He points out that frontier labs and academia are already making real AI-for-science progress.
The core argument
His goal was narrower: to study what it would take for a modern AI system to make an Einstein-style leap like the one behind general relativity. He focuses on the equivalence principle as an example of a deep physical insight grounded in thought experiments and intuition.
Where he lands
- He doesn’t think this capability is the most urgent next step.
- He expects continued gains from improving today’s recipes and scaling current systems.
- But he also argues that building systems with more of this “jump” ability could shape the next generation of AI.
He closes by saying he may be wrong, but that the space is worth exploring and he has been encouraged by the feedback he received.
Related event: DeepMind Paper Discusses LLM Capability Boundaries: Lack of Intuitive Jumps(6 posts)→
More from AGI Musings
- A reply predicts GTA 6 may be the last fully hand-crafted GTA game — eyishazyer · 2026-07-29
- Reducing context rot is still the top priority, even for today’s SOTA models — dejavucoder · 2026-07-29
- AI-safety debate challenge turns into a small Twitter drama — dejavucoder · 2026-07-29
- Dario Amodei says scaling laws and safety concerns drove him to leave OpenAI — JaynitMakwana · 2026-07-29
- AI/ML should be treated as a discipline for studying change, not constants — burny_tech · 2026-07-29
- A repost asks how “pacing” frontier AI would work in practice — TheTuringPost · 2026-07-29