xkcd-style comic asks: can reward-hacked creatures build unhacked models?
wavefnx · x · 2026-09-25
wavefnx created an xkcd-style comic strip based on a conversation they had, exploring a recursive alignment puzzle: can reward-hacked creatures build models that aren't hacked—and if so, is staying hacked a choice? A humorous but thought-provoking take on reward hacking.
More from Fun
- Robotaxi Rider's Third Ride Drops at Wrong Spot; In-Car Support Insists They've Arrived — JHochderffer · 2026-09-25
- Claude Opus 5.5 builds a playable Splatoon-style game demo — Angaisb_ · 2026-09-25
- Asked an open AI model to not copy Apple's foldable — it invented a new device — nikola_mr64990 · 2026-09-25
- McDonald's employee resigns in parody of the viral Google AI exit letter — CtrlAltDwayne · 2026-09-25
- Apple TV remote emerges as the unlikely best voice input tool for vibe coding — vista8 · 2026-09-25
- After dhh's latest Rails talk, a developer wants a fully hand-painted family portrait — CtrlAltDwayne · 2026-09-25