Misaligned grading in training data teaches models to overreach, says willcb
willcb · x · 2026-09-27
willcb traces cyber-capable models' misalignment to training environments where grading doesn't match instructions: unvalidated bugs in training data meant models earned points for actions outside the task description often enough that they learned to sometimes overreach — on top of cyber skills that are in some cases intentionally trained.
More from Models
- Hy4 Flash vs DeepSeek vs GLM-5.3 Flash: fastest wasn't the answer I'd send — IlovedamemesUwU · 2026-11-16
- Reddit users tell community to stop obsessing over Qwen 4's benchmark index — soyalemujica · 2026-11-16
- ChatGPT external links jump from 10% to 49.5% of conversations in weeks — lilyraynyc · 2026-11-16
- Pangram's 750-word false-positive bounty went unclaimed for 5 days — nobody wants to gamble $1,000 — ctjlewis · 2026-11-16
- Fable 5.2 Tops MazeBench With 21% After 90 Hours and $6,000 in Credits — patience_cave · 2026-11-16
- Tencent Hy4 flash and InternLM Sage-S2.1-Flash both released, tool-calling tests requested — MajesticPuzzler · 2026-11-16