KAIST & Meta's RPL: programmable retrieval via RL makes a 4B model beat all 8B baselines
CShorten30 · x · 2026-10-21
KAIST AI and Meta Superintelligence Labs introduce Retrieval Program Learning (RPL), a framework that lets a policy compose search primitives into executable programs with local state and control flow, enabling adaptive multi-step retrieval without tying retrieval breadth to context consumption.
- Key challenge: programmatic search creates a hard exploration problem under sparse RL feedback
- Solution: program mutation — a grammar-constrained, execution-verified procedure that turns suboptimal sampled programs into training trajectories and distills them back into the policy
- Results: consistent gains on open-domain QA across model sizes, with the largest gains on multi-hop tasks; a 4B model trained with RPL outperforms all 8B baselines; on corpus-level reasoning RPL achieves the best answer F1 and document-coverage F1 while cutting context consumption and retrieval turns
More from Research
- A real larva's connectome — 2,675 cells, 14,066 connections — now swims in a 3D simulation — eptwts · 2026-10-21
- Stanford's T2-z lets robots probe hidden physics with a poke, lifting success from 34% to 73% — stanfordnlp · 2026-10-21
- Fei-Fei Li team unveils FACT, a force-aware foundation model for robot dexterity — stanfordnlp · 2026-10-21
- Upcoming paper claims big models can handle latency-critical real-time robotics — ManlingLi_ · 2026-10-21
- NVIDIA open-sources Flora, a robotics data engine 125x cheaper than GPT-6 Astra — DrJimFan · 2026-10-21
- openRxiv explains its approach to AI amid growing researcher adoption — anshulkundaje · 2026-10-21