Mechanistic Interpretability Might Finally Settle the Chomsky vs. LLM Debate

mathemagic1an · x · 2026-08-06

The author revisits Noam Chomsky's post-ChatGPT critique in the NYT, where Chomsky argued that LLMs cannot provide "correct explanations" of language. The poster counters that correct explanations can indeed be learned by marinating in big data.

The thread suggests that the debate over the nature of language will ultimately be put to bed by advances in Mechanistic Interpretability.

Original post →

Mechanistic Interpretability Might Finally Settle the Chomsky vs. LLM Debate

More from AGI Musings