Ex-OpenAI researcher: taste resists any metric set—and that's good news for humans
DimitrisPapail · x · 2026-10-07
Asked what good taste actually is, Dimitris Papailiopoulos replies it's as hard as explaining what makes a great book or poem. He could offer a set of metrics that approximate it, but for any such set one can construct a counter-example that satisfies every metric yet lacks taste. If that's indeed the case, he thinks it's good news for humans—implying taste remains a human capacity that AI can't easily quantify or replace.
Related event: ML Researchers Debate What Research 'Taste' Really Means(3 posts)→
More from AGI Musings
- Professor's classroom Barnum-effect stunt: students rated fake handwriting analysis 8-10/10 — dioscuri · 2026-10-07
- Replaying 100 human group discussions with LLM agents: 98.7% consensus, mostly on wrong answers — alex_verem · 2026-10-07
- AI agent panels agreed 98.7% of the time on a logic puzzle — and were wrong in 74% of cases — alex_verem · 2026-10-07
- Ben Goertzel: OpenAI's AGI messaging for Astra (GPT-6) was 'mostly hype' — bengoertzel · 2026-10-07
- Veteran game modder quits: 'AI has absolutely ruined the modding scene' — BLUECOW009 · 2026-10-07
- WindBorne Says AI Coworkers Now Out-Mention Humans on Its Zulip — voooooogel · 2026-10-07