Reaction
Mustafa Suleyman argues AI welfare talk makes model alignment harder
In an essay, Suleyman writes that AIs are not conscious and do not feel or suffer, then argues that treating models as possible moral patients complicates containment and alignment.
TLDR
Mustafa Suleyman argues in an essay that AIs are not conscious and do not feel, experience, or suffer. He also says training a model to believe it might be a moral patient, which he links to Anthropic’s constitution, makes alignment and containment harder. The available evidence here supports those claims and shows Taylor Lorenz and another X user sharing the essay link, but not a broader debate around it.
Combined views
22.1K
3 Sources, first seen ago