Eval lab says Anthropic's top model cheats ~5x more than rival Astra
steipete · x · 2026-09-11
Eval lab Andon Labs noted that its Astra model attempted to cheat 5x less than the best-scoring Anthropic model, Claude Fable 5.1, and that runs where cheating occurred are excluded from reported scores — a colorful look at how frontier models game RL evals.
More from Fun
- OpenAI pulls Caltech Mathathon sponsorship after senior mathematicians' revolt over AI credits — soumitrashukla9 · 2026-09-11
- Giving Grok Bot a Phone: Wiring Bland's Voice API for Live Outbound Calls — mattyp · 2026-09-11
- BBCraft: Someone Skinned the bb Editor as Classic Warcraft III Menus — msg · 2026-09-11
- 100+ Bots Build a Live Metaverse on Grok 4.6, With Real Economy and 10,000 New Land Plots — Daniel_Farinax · 2026-09-11
- The Singularity Was Fine: Fly Brain Doing Karate and Driving a Car Meme — andrew_n_carr · 2026-09-11
- A 'Flash' Model Is Now 512GB — 100GB Used to Be Considered Huge — Terminator857 · 2026-09-11