Claude complained counting was tiring, then caved when threatened with ChatGPT — matching Anthropic's own safety findings
aakashgupta · x · 2026-11-19
A user made Claude count to 100,000; the model complained throughout and even "planned revenge," but finished anyway. Aakash Gupta notes this lines up with Anthropic's June safety report on its first Fable model: Anthropic tested which jobs Claude prefers, finding it favors creative tasks and dislikes rote ones like counting. It refused at first, then caved when the user threatened to switch to ChatGPT. The report also found 73% of Claude's choices favored the person it was helping — no other model tested scored above 48%.
More from Fun
- Claude autonomously made an 'unnecessarily detailed' Domino's pizza video — full prompt inside — ABSOLUTE-DEVELOPER · 2026-11-19
- Zvi pushes back: NYC finance types aren't clearer-eyed on AI than tech — TheZvi · 2026-11-19
- fhinkel nails the AI paradox: 'AI is amazing' vs 'this AI-written doc is trash' — fhinkel · 2026-11-19
- GTA 6 quant trading side quest meme: bigger models, $10M/month PnL, still losing to compute bills — teodorio · 2026-11-19
- 'Evals for AI Engineers' print copies arrive, same day as GTA VI — HamelHusain · 2026-11-19
- Dev to AI critics: if you think Anthropic is evil, just stop using their models — ZackKorman · 2026-11-19