Reinforcement learning research

• The Teacher Is a Direction, Not a Destination: Extrapolating RL-Induced Representation Residuals in On-Policy Distillation

  RIDE pushes a student model beyond its reinforcement-learning teacher by training student hidden representations to overshoot teacher representations along the layerwise residual vector between the teacher and its pre-learning base model, bypassing noisy output-token sampling.

  2026-09-29 · Reinforcement learning · Language models

• Scaling Properties of Same-Family On-Policy Distillation

  By measuring student learning progress through divergence from initialization, this paper fits joint power laws that predict a student language model's peak test accuracy and learning speed from model scale and teacher accuracy before distillation begins.

  2026-09-26 · Language models · Reinforcement learning · Efficient AI

• False Frontiers: Diagnosing and Mitigating Co-Cheating in Self-Evolving Search Agents

  CrossFit partitions source documents into separate groups called folds and scores questions using a solver trained on complementary folds, reducing same-source error feedback to lower false agreement and improve search accuracy.

  2026-09-30 · AI agents · Reinforcement learning · Evaluation and benchmarks