AI news roundup • Anthropic uses Claude agents to red-team its monitors #AI #Anthropic #AISafety
- 1 réponses
- 0 partages
- 0 j'aime
- 2 vues
AI news roundup • Anthropic uses Claude agents to red-team its monitors #AI #Anthropic #AISafety
Anthropic set thousands of Claude agents to sneak misaligned behavior past a training-data monitor using harmless-looking examples, which can now be used to build better monitors. t.co/SP1qvEiMqA #AISafety #Claude