Emergent wisdom in language models and the limits of AI decentralization

Dalrymple says a language model can understand what is right without reliably acting on that understanding. He argues against a single controller for Earth, favoring multiple autonomous jurisdictions. A reply on X questions whether the possibility of an “evil” direction makes emergent wisdom reassuring and argues for empowering people who can be trusted with power, rather than pursuing decentralization for its own sake.