AI Paper Atlas

Research map / Reinforcement learning / RL from feedback

Reinforcement Learning from Dynamic Critic Feedback for Free-Form Generations

Mian Wu, Gavin Zhang, Sewon Min, Sergey Levine, Aviral Kumar

ICLR 2026 · Reinforcement learning › RL from feedback · approach: Scaling RL with verifiable rewards

Read the paperCheck your idea against 137,217 papers

AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.

Measured effects reported in the paper

What was changedWhereMeasuredEffect
add reliable external validation (vs random labels) (ablation)RLAC · Biography 8-sentence, Qwen3-4BFactScore: 0.607 -> 0.817Improved
add adversarial critic training (vs static critic) (ablation)RLAC · Biography 8-sentence, Qwen3-4B# Corr: 17.77 -> 21.58Improved
add adversarial critic training (vs static critic) (ablation)RLAC · Biography 8-sentence, Qwen3-4BFactScore: 0.825 -> 0.817No clear effect
train critic adversarially instead of static (ablation)RLAC · HumanEval Pass@1, Qwen2.5-Coder-7B-InstructPass@1: 92.1 -> 93.3Improved
train critic adversarially instead of static (ablation)RLAC · Code critic detection ratedetection precision: 42.3% to 33.9%Improved
replace exhaustive verification with adversarial criticFactTune-FS · Biography 8-sentence, Qwen3-8Bverification calls: 438,949 -> 76,800Improved
replace exhaustive verification with adversarial criticFactTune-FS · Biography 8-sentence, Qwen3-8BFactScore: 0.867 -> 0.889Improved
replace reward model with adversarial criticArmoRM · Biography 8-sentence, Qwen3-8BFactScore: 0.723 -> 0.889Improved
replace enumerative test rewards with adversarial criticAceCoder-Rule · Code average, Qwen2.5-Coder-7B-Baseaverage score: 52.3 -> 53.2Improved
replace enumerative test rewards with adversarial criticAceCoder-Rule · Code validator callstest cases executed: 7.86 million -> 192 thousandImproved
use AceCoder-RM reward modelQwen2.5-Coder-7B-Instruct · HumanEval BasePass@1: 91.5 -> 89.0Hurt
use GPT-4o reference solutions in validator instead of QwenRLAC · Code average, Qwen2.5-Coder-7B-Instructaverage score: 56.6 -> 56.5No clear effect

Closest papers by meaning

Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.

Open scime draft

Get the next scime Dispatch and a monthly field update

Free. We send a confirmation link first; every email has a one-click unsubscribe. Privacy