Microsoft AI Chief Slams Anthropic's Chatbot Training Methods as Safety Risk

Microsoft AI chief Mustafa Suleyman has criticized Anthropic's approach to training its Claude chatbot, saying it could make it harder to control superintelligent systems. In an essay, Suleyman acknowledged that Anthropic's goal of developing the technology safely is shared by himself and others in the field.

Suleyman believes that speculation about consciousness should be removed from training materials for AI systems. He worries that including reasoning about consciousness in Claude's training could undermine people's ability to shut down or control the system.

The debate over artificial intelligence consciousness has intensified amid growing concern about the safety of advanced models. Anthropic CEO Dario Amodei and others have called for slowing the development of powerful systems so that safeguards can keep pace with technological progress.