Amos Tversky

A Nature analysis sorts LLM human proxies into four roles that each need their own validity test

The analysis argues that a model standing in for people can be sound for one research purpose and unsound for the next, and that the four purposes it identifies each rest on a different claim about human likeness.