Tag
This paper proposes the EDV framework, which uses multiple heterogeneous agents in execute-distill-verify stages to build reliable experiences for LLM agents, preventing self-confirmatory errors and improving performance on long-horizon benchmarks.