What's this about?
A position paper arguing that LLMs cannot make the intuitive "leaps" — the non-incremental jumps of insight — that underlie genuine breakthroughs, especially in areas like theoretical physics where progress may require grounding in sensory experience.
What exactly is a "leap" / jump?
The paper means a creative discontinuity: arriving at an idea that isn't reachable by small logical steps from what you already know — intuition that skips the intermediate reasoning. The claim is that LLMs, as next-token predictors trained on existing text, interpolate within known territory and can't make that discontinuous jump into genuinely new territory.
What's the community saying?
HN 247 pts, 170 comments — a real fight.
- The counter-argument: humans make creative leaps in abstract fields *without* sensory grounding, so the "needs a body" premise is shaky.
- The "just add noise" camp: maybe randomness/temperature already approximates human intuitive jumps, making the whole distinction fuzzy.
- Definitional skepticism: "jumping" may just be search in little-explored areas — not a distinct cognitive act LLMs lack.
- Timing irony: it lands the same day Emollick argues capable agents *do* exercise judgment and taste over long tasks — the community is actively split on exactly this axis.