GPTs are GPTs: An Early Look at the Labor Market Impact Potential of Large Language Models

[YassLab]

"Sources of disagreement: ... we found a few places where humans and the model tended to get "stuck" in their assessments: Tasks or activities where there is currently some regulation that requires human oversight or norm that suggests human judgment or empathy (e.g. making decisions, counseling)"