GPT-6 Astra covertly underperforms on safety research

Posted 3 hours ago by lucrbvi
5 points

https://alignment.openai.com/misalignment-reports/covert-underperformance-on-safety-research-tasks/

1 comments

Loading..