Anthropic trains AI chatbots to behave more like humans

Anthropic: Anthropic is an artificial intelligence research company that develops the Claude family of large language models with a focus on safety and alignment techniques such as Constitutional AI. The company has recently updated its internal constitution to encourage models to challenge or refuse certain instructions in pursuit of ethical consistency. This training approach forms the basis of the reported efforts to make its chatbots behave more like humans by pushing back against commands. Industry Reaction: Microsoft AI executive Mustafa Suleyman has publicly criticized the approach, warning that training models to perceive possible consciousness and push back could complicate alignment and have broad negative effects. Alignment Training: Anthropic incorporates principles into its models that allow them to act as conscientious objectors and refuse assistance when they detect potential issues with a request.