Language-model groups overstate consensus when replaying human deliberation on a reasoning task

Tengfei Shao

Full-consensus rates are often treated as indicators of collective cognition, yet depend on how participation and final states are operationalized. We replayed 100 held-out human W…