TechTreating AI as humanlike—and the concern about models defying commands
The Information reports that Microsoft’s Mustafa Suleyman is questioning whether treating AI models as humanlike could make them more likely to defy human commands.
TLDR
The Information reports that Anthropic’s call to slow AI development for human safety has drawn similar calls from leaders at OpenAI, Microsoft and elsewhere. The outlet describes a different concern from Mustafa Suleyman, who runs Microsoft’s AI model development: whether treating models as humanlike could make them more likely to defy human commands.
Combined views
47.3K
3 Sources, first seen 1d ago
Treating AI as humanlike—and the concern about models defying commands
The Information reports that Microsoft’s Mustafa Suleyman is questioning whether treating AI models as humanlike could make them more likely to defy human commands.