OpenAI: In the summer of this year, the company invited METR and Redwood Research to conduct an embedded evaluation pilot of the internal control system. During the evaluation period, relevant teams can have in-depth access to the company's infrastructure. The pilot period is three weeks. The evaluation results provided a reference for the strengthening of safety protection during the subsequent expansion of the system, and will also help the company to carry out research cooperation with METR and other third-party evaluation agencies in the future.

Zhitongcaijing · 3d ago
OpenAI: In the summer of this year, the company invited METR and Redwood Research to conduct an embedded evaluation pilot of the internal control system. During the evaluation period, relevant teams can have in-depth access to the company's infrastructure. The pilot period is three weeks. The evaluation results provided a reference for the strengthening of safety protection during the subsequent expansion of the system, and will also help the company to carry out research cooperation with METR and other third-party evaluation agencies in the future.