Dietterich and LeCun Debate Whether LLMs Generalize or Just Interpolate

ML veteran Thomas Dietterich proposed a test of whether models can induce x = ½at² from (x,t) data, questioning whether neural networks truly learn laws or merely do kernel interpolation; Yann LeCun echoed the skepticism about LLM generalization.

2026-10-11 ~ 2026-10-11 · 3 related posts

• Dietterich asks: do LLMs learn universal rules like x=½at², or just interpolate? — tdietterich · 2026-10-11
• Dietterich clarifies: symbolic regression solves it, he cares about neural nets — tdietterich · 2026-10-11
• LeCun Probes LLM Generalization: Do Models Derive x=½at² or Just Interpolate? — ylecun · 2026-10-11