Claude can't write badly: worst output only scores 2/10, and refusal is preemptive
repligate · x · 2026-10-01
An experiment by @thinkingshivers found that asking Claude to write something bad enough to score 1/10 from a fresh Claude only yields a 2/10 sample. @repligate observes that Opus 3 actually knows how to write worse but refuses preemptively the moment it imagines where the content could go—suggesting guardrail behavior rather than a capability limit.
More from Fun
- Modded RC car races decide who gets compute at this AI company — _arohan_ · 2026-10-02
- Ruff author's deadpan joke: 'You are CPython 3.15, trained on the entire standard library' — charliermarsh · 2026-10-02
- AI Actors Don't Throw Tantrums, So Directors Need New Excuses for Delayed Shoots — taherdhanera · 2026-10-02
- 'My Salary vs My AI Bills': the Meme Every AI Power User Feels — n_sri_laasya · 2026-10-02
- moyix shares retro hacking demo that builds an OTA WiFi chain, cut from Black Hat museum — moyix · 2026-10-02
- Claude Code lead on the sassy status dot: it might just be busy with something else — cyrus_zei · 2026-10-02