When Will LLMs Produce New Theories?

burny_tech · x · 2026-07-19

A post relayed a perspective on the upper limits of future LLM capabilities: if a model could one day propose a "new theory" in fields like the Geometric Langlands Program—approaching the 2024 work of Gaitsgory and Raskin—it would represent a massive capability leap. The author estimates this is still about two years away.

Related event: Debate Sparks on LLM's Potential to Generate New Theories(2 posts)→

Original post →

More from AGI Musings

AGI Musings channel →