1 paper
Andrea Wynn, Harsh Satija, Gillian Hadfield
While multi-agent debate has been proposed as a promising strategy for improving AI reasoning ability, we find that debate can sometimes be harmful rather than helpful. Prior work…