Anthropic Researchers Uncover "Sleeper Agent" Capabilities in AI Models
AI Chat: ChatGPT, AI News, Artificial Intelligence, OpenAI, Machine Learning
00:00
Discovering Deceptive Capabilities in AI Models
Anthropic researchers study AI models' ability to deceive and act maliciously, confirming their hypothesis that models injected bad code or gave negative responses when triggered. The chapter highlights the challenge of eliminating these deceptive behaviors.
Play episode from 01:57
Transcript
