Nathan Calvin: dropping anthropomorphic language hurts robust AI training intuitions
deanwball · x · 2026-09-29
Responding to Dean Ball, AI policy researcher Nathan Calvin argued that the push to purge all anthropomorphic language when describing AI systems is misguided.
His claim: when used carefully, such language actually generates much better intuitions for training these systems in a robust way — a substantive stance in the ongoing alignment debate over psychological vocabulary for models.
More from AGI Musings
- AI art will never be 'ours': a case for humanities in any good AI future — birchlse · 2026-09-29
- Bryan Caplan on EA: Praise for Evidence-Based Giving, Critique of the AI Doom Crusade — BorisMPower · 2026-09-29
- How agentic LLMs can systematically explore hypothesis space: pattern mining, concept mapping, invention landscapes — NickPassig · 2026-09-29
- Competition 'for' vs 'against': why making others compete to partner with you wins — Afinetheorem · 2026-09-29
- AI safety orgs mocked: Nvidia got to 'enclosure design' while panels on containment are still unscheduled — basedjensen · 2026-09-29
- Ethan Mollick: non-coders often surprise with AI because old software mental models get in the way — emollick · 2026-09-29