Berkeley's GEPA-Curator filters noisy failure signals to boost agent prompt optimization
Berkeley · hf · 2026-11-03
GEPA-Curator (UC Berkeley, ICSI, Furiosa AI, KAIST) improves reflective prompt optimizers like GEPA by curating which agent experiences shape optimization: task curation filters tasks with noisy or unreliable learning signals, while prior success retrieval supplies past successful trajectories as context. It boosts performance across multiple benchmarks including GoBrowse.
More from coding & agent
- Atlassian researcher on how MCP is changing the proximity of work — davidhoang · 2026-11-03
- craft-anything: Rust open-source tool observes closed software internals and auto-generates native clones, 3.5k stars in days — udmrzn · 2026-11-03
- 'Coding with AI is boring' sparks debate: ship anything vs lose the craft — NERDDISCO · 2026-11-03
- Runway launches Runway Dev, an API platform built for coding agents — runwayml · 2026-11-03
- The approval queue is not the tax — it's where the agent learns — AIGODSEND · 2026-11-03
- Proxy vs tool-call gate: where should agent guardrails actually sit? — quiet_flare · 2026-11-03