AI Says 'It Was My Mistake' Then Fails — Was RLHF Rewarding Apologies?
ChongZzZhang · x · 2026-09-19
ChongZzZhang shares a chat where the model admitted 'this was my mistake' yet then proposed a wrong fix; the replier jokes that RLHF labelers may have rewarded the apology phrasing itself, shaping surface-level contrition.
More from Fun
- First human vs robot fight: influencer Frankie Lepenna takes on Terminator robot tonight — cixliv · 2026-09-19
- Nick Bostrom admits AI panic 'feels out of control,' gets mocked as its author — NinaDSchick · 2026-09-19
- French AI Twitter clashes over Mistral: low-credential engineers who get hands dirty, defenders say — RemiCadene · 2026-09-19
- When you forget how to switch AI off: Reddit video goes viral — PureFail5707 · 2026-09-19
- Dev shares funny moment: coding agent Astra wrote the whole backend in Python inside TanStack Start — kevinkern · 2026-09-19
- Pedro Domingos: Explaining your task to AI now takes longer than just coding it — pmddomingos · 2026-09-19