Neel Nanda vs ICML paper: is anthropomorphizing LLMs a useful lens?

sebkrier · x · 2026-09-04

A debate over anthropomorphic language for LLMs: Neel Nanda argues models trained on trillions of human tokens naturally reach for human abstractions, making anthropomorphism a useful interpretive lens — not a consciousness claim. An ICML position paper by Xin Cynthia Chen counters that anthropomorphized concepts are imprecise and carry unexamined assumptions like a "consistent persona with intent," calling for more scientifically careful attribution language.

Related event: Neel Nanda Defends Anthropomorphizing Language Models(3 posts)→

Original post →

More from AGI Musings

AGI Musings channel →