One warning sentence helps AI agent groups with incorrect context
The warning recovers a maximum of 0.18 accuracy with voting.
In tests with 3 open-weight models, incorrect text removed 0.12 to 0.43 accuracy from one agent. Debate with 1 of 3 agents that read it gave the same result as a vote. A warning of one sentence recovered a maximum of 0.18 accuracy with voting, as the authors show.
Sources
Posted
