swyx on AI doom-posting: no penalty for being wrong, and Gemini's RL team printed the bearish take as motivation
gleech · x · 2026-11-20
swyx points out the incentive structure of AI Twitter: you can post "odds of reaching SOTA have dropped to zero," farm likes, then post an apology form months later when the target actually reaches SOTA — and get even more likes. "There's no consequences at all for inaccuracy and every incentive for hyperbole."
The punchline: a member of Google's Gemini RL scaling team confirmed they printed the SemiAnalysis piece and pinned it on their desks as daily motivation — "but the times they are changing."
More from Fun
- Mad Men still mistaken for a GTA 6 open-world mod — dreamwieber · 2026-11-20
- ChatGPT, asked who wins the AI era, picks Google over its own maker OpenAI — tinyfool · 2026-11-20
- The Hundred-Page Machine Learning Book on sale for $14 (from $40) on Leanpub — burkov · 2026-11-20
- three.js team unveils Neon Scramble, turning Tokyo's Shibuya Crossing into a browser parkour game — AI_Andrew · 2026-11-20
- How many marriages will survive when AI can verify every argument? The divorce industry's big decade — VraserX · 2026-11-20
- User jokes: 'If Jensen doesn't come back, I'll leave Nvidia' — Yamapama · 2026-11-20