User data still matters at small scale, but mining it efficiently may require strong models first
teortaxesTex · x · 2026-11-21
Follow-up: at very small scales and low power user share — like at Chinese labs — user data still holds value since the task distribution isn't even known yet. But it's a chicken-and-egg situation: perhaps you need very strong models to begin mining user data efficiently.
More from AGI Musings
- Jab: Chinese AI Decision-Makers Outside the Labs Are 'Binge-Watching Marcus and Zitron' — teortaxesTex · 2026-11-21
- Peter Diamandis: the way to public AI trust is solving daily problems — PeterDiamandis · 2026-11-21
- Gary Marcus Shares a Jab at Dwarkesh Over Bad AI Predictions — GaryMarcus · 2026-11-21
- FT data: retirees now out-consume working-age people across six countries — isnit0 · 2026-11-21
- Google AI chief: Gemini already helps build the next Gemini — VraserX · 2026-11-21
- Researchers warn even eminent minds underestimate AI's exponential pace, risking policy missteps — satnam6502 · 2026-11-21