多智能体

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

Full-consensus rates are often treated as indicators of collective cognition, yet depend on how participation and final