ShadowBench tests if frontier models can recreate 150-year-old shadow puppets
tarantulae · x · 2026-11-06
spratyey introduces ShadowBench, a playful benchmark testing whether frontier models can recreate shadow puppets. The model gets an RL environment with a curtain and 3D hands, and must pose them joint by joint to match a diagram from a 150-year-old puppet book, crafted at perpetualsi. Commenters quip that we may finally have run out of benchmarks to implement.
More from Fun
- AI startup founder publicly trashes Azure Portal UI, tags Nadella directly — PaulYacoubian · 2026-11-06
- Wired-up Gmail meets AI mining game: clear your inbox, dig for the Crown of Wonder — omooretweets · 2026-11-06
- Police department jokes that arrestees' one phone call is now Microsoft Teams — Kyrannio · 2026-11-06
- AI-generated Willem Dafoe Green Goblin shots show how far image models have come — egeberkina · 2026-11-06
- Benchmarking which AI lies best, using a roleplaying game — ownworldman · 2026-11-06
- Naming your sieve variable 'virgins' leads to an AI suggesting a meeting with the pastor — akbirthko · 2026-11-06