AI 导读
据 WSJ 报道，三名因涉嫌不当行为被 OpenAI 解职的安全研究员致信董事会，要求停止任何进一步削弱人类监控 AI 推理能力的开发。其中 Tomek Korbak 和 Mikita Balesni 曾主导 2025 年指出链式推理监控有用但脆弱的论文。OpenAI 则表示解职原因是未妥善处理敏感信息。
正文
WSJ: Three safety researchers fired by OpenAI for alleged misconduct have asked its board to halt any development that further reduces humans’ ability to monitor AI reasoning.
Two of them, Tomek Korbak and Mikita Balesni, led the 2025 paper that warned chain-of-thought monitoring is useful but fragile.
OpenAI however says the dismissals concerned mishandled sensitive information.
来源：Rohan Paul · x.com