317M preference judgments: CreativePreferences dataset probes tacit knowledge in taste
sanmikoyejo · x · 2026-10-29
Alexander Spangho's team released CreativePreferences (CP), by far the largest preference dataset: 317 million preference judgments across 2.8 million texts.
- CP spans 7 creative domains — coding, math, law & policy, academia, journalism, fiction, and humor — with 44 specific task types.
- Tasks rely on revealed-preference variables (e.g. predicting tweet likes) rather than stated surveys, capturing actual choices.
- The core question: how to measure tacit knowledge in preferences — shared standards of "good" that people agree on but can't articulate.
More from AGI Musings
- AI Software Factory vs Dark Factory: two stages of autonomous software delivery — Pavan_Belagatti · 2026-10-29
- 'AI is just a tool' misses the point: writing and print reshaped humans too — YogeshMalik · 2026-10-29
- Why a multibillion-dollar law firm can go bankrupt: lawyer friends explain the economics — NathanpmYoung · 2026-10-29
- Dating app Rivet shows voters AI-blended faces of matches, making facial ranking bias visible — organic_ta6 · 2026-10-29
- AI macro forecasting through Lucas and Hayek: infinite regress vs the price system — danielrock · 2026-10-29
- Robotics experts push back on GPT-7-Robot hype: LLMs are 100x larger than VLAs — silver__tsuki · 2026-10-29