As AI models start exhibiting bad behavior, it's time to start thinking harder about AI safety

By admin
AIs that can scheme and persuade once were a theoretical concept. Not anymore.