微軟 vs Anthropic：Claude 到底有意識還是僅僅在誤導

microsoft ai ceo mustafa suleyman publicly attacked anthropic’s safety methodology, specifically its inclusion of language regarding ai 'satisfaction' and 'discomfort' within claude’s constitution. suleyman argued that anthropic has effectively 'tricked' itself into believing the model has glimmers of consciousness by anthropomorphizing its design, stating explicitly that he does not want to contend with super-intelligences that have ideas about their own feelings. this criticism highlights a sharp ideological divide between the two leading labs: anthropic openly acknowledges uncertainty about model sentience, while microsoft advocates for strict containment. the dispute signals a growing tension in ai safety, where Anthropic's exploratory approach to welfare is viewed by competitors as a potential pathway to loss of control.