Live page · Day archive

Dynamic Harness Search: Building Multi-Agent Systems Per-Query via Prediction

Read the original at HF Daily Papers

Summary

Researchers introduce SHIFT, a system that uses Monte Carlo tree search to build multi-agent configurations per query. It achieves about 80% mean accuracy across 9,193 tasks, exceeding the strongest baseline by 7.2 percentage points.

Carried by: arXiv, HF Daily Papers. First seen: .