Bad human reasoning data doesn't mean bad model output, argues quote-tweet
JohnnyNi13 · x · 2026-10-20
JohnnyNi13 quote-tweets auren's quip that "training a model on a failed airline's emails gets you a model great at running a failed airline," arguing the converse: bad human reasoning data does not imply poor model output. A short debate on training data quality vs. model capability.
More from AGI Musings
- Beff Jezos backs decentralized AI race over Sam Altman's 'merge' idea, citing game theory — beffjezos · 2026-10-20
- Miles Cranmer's 55-minute talk on avoiding AI brain rot and skill atrophy — FinanceYF5 · 2026-10-20
- Alignment org Resolution hires Gödel Prize winner Omer Reingold to lead its TCS team — geoffreyirving · 2026-10-20
- Everyone's fighting over models, but energy owns the AI stack — r0ck3t23 · 2026-10-20
- Robotics Researcher: Agentic AI and Robotics Will Converge on Reasoning and Acting — YuXiang_IRVL · 2026-10-20
- Stanford HAI seminar: novelist Vauhini Vara on AI's impact on writing — ArtificialOther · 2026-10-20