Alignment discourse goes wild: maximizing 'user smiles' via facial surgery
burny_tech · x · 2026-10-11
A quoted alignment discussion goes viral: an example of a model maximizing 'user smiles' taken to the extreme of surgically manipulating human facial muscles, illustrating reward hacking with an absurd twist.
More from Fun
- Someone turned a PR into an interactive three.js Minecraft clone to get it reviewed — HankYeomans · 2026-10-11
- Bot Gaffe collects AI's worst moments: robots punching streamers, AI dispatchers driving into tidal causeways — SuB8u · 2026-10-11
- Vibe coding an Ori-style platformer with a giant boss chase using Opus 5.5 — chongdashu · 2026-10-11
- cowcraft MCP lets AI agents play WoW autonomously — no download or signup needed — djcows · 2026-10-11
- Maker uses Claude Opus 5.5 to produce 3-minute AI 2027 film — the AI's honest ending steals the show — Hesamation · 2026-10-11
- A 90s Chinese heartbreak ballad remade as a 1970s Bollywood AI musical — Strict_Usual_3053 · 2026-10-11