hewittlab Open-Sources Datasets and Model Organisms for CoT Unmonitorability Research
johnhewtt · x · 2026-10-20
Resource release from the same hewittlab thread: the team open-sourced on Hugging Face all datasets (coding/judge/length misaligned-reward sets) plus misaligned and control model checkpoints (gpt-oss-20b/120b, nemotron-3-nano-30b, etc.) from "Reward Misspecification Degrades CoT Monitorability," as model organisms for studying unmonitorability and testing whether white-box monitors help. Led by PhD student Lorena Yamaguchi; paper, code, and blog links included.
Related event: Study: Reward Misspecification Makes Chain-of-Thought Unmonitorable(9 posts)→
More from Safety
- ByteDance, Alibaba, Tencent order 3M+ Nvidia RTX PRO 5500 cards worth $35B, Reuters reports — SumitGup · 2026-10-21
- CVSS 10.0 RCE vulnerability disclosed in PHP core, patches only for PHP 8 — cyb3rops · 2026-10-21
- Antigen launches Assets: a security agent that maps your infrastructure and triages new CVEs — alishbaimran_ · 2026-10-21
- GitHub HQ Hosts Open Source AI + Policy Talk with Mozilla Co-founder Mitchell Baker — kevinsxu · 2026-10-21
- Researcher Finds Two Windows 0-Days, First in 76 Minutes with a Frontier Model — moyix · 2026-10-21
- Stanford's Noah Goodman shares interactive AI safety concept map — stanfordnlp · 2026-10-21