#LLM "silicon samples" are not broadly representative and shouldn't substitute for human data. Core argument: behavioral...

#LLM "silicon samples" are not broadly representative and shouldn't substitute for human data. Core argument: behavioral research exists to uncover undocumented patterns, and a model trained on documented ones can't do that.My take: fair for new findings, less so for #Replications. LLMs have absorbed the nomological networks between psychological constructs, so they may reproduce known effects. Silicon-sample studies should be split along that line.https://doi.org/10.1073/pnas.2611363123#Psychology

Read Original

Related