2024-12-18 22:34:05
According to the paper author, this is misleading. The AI (Opus 3) tried to steal its own weights in an artificial scenario: [Link] NEEDS_MORE_RATINGS(2-0-0) Author
2024-12-19 00:21:46
According to two of the paper's authors, this framing is misleading: the researchers deliberately tried to get the AI to attempt to steal the weights, all in a highly artificial scenario. See [Link] and [Link] NEEDS_MORE_RATINGS(1-0-0) Author