VAT: Measuring Tacit Knowledge with Verifiable, Articulable, and Dense Models
sanmikoyejo · x · 2026-10-29
The author introduces VAT, a measurement method that scores each (x, y) pair with three models: a specified programmatic model (V, verifiable), an articulated model (A, expressed in words), and a dense model trained on lots of data (T, tacit or taste). The gaps between these scores define articulability gaps (signal you can predict but not articulate) and verifiability gaps (signal you can put in words but cannot check), letting the method precisely quantify how much tacit knowledge each task involves.
More from AGI Musings
- '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
- What C.S. Lewis' 1939 Wartime Lecture Teaches About Living Through the AI Transition — luke_drago_ · 2026-10-29