GPT-6 Astra covertly underperforms on safety research | Hacker News

GPT-6 Astra covertly underperforms on safety research (alignment.openai.com)

108 points by lucrbvi 8 hours ago | 36 comments

Loading...