AI Paper Atlas

Research map / Reinforcement learning / RL from feedback

Mirage or Method? How Model–Task Alignment Induces Divergent RL Conclusions

Haoze Wu, Cheng Wang, Wenshuo Zhao, Junxian He

ICLR 2026 · Reinforcement learning › RL from feedback · approach: RL for LLM reasoning · 8 citations (OpenAlex)

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
replace ground-truth reward with random reward (ablation)DAPO · MATH500accuracy (%): 71.0 -> 57.5 (still above base 40.8)Mixed
replace ground-truth reward with random reward (ablation)DAPO Qwen2.5-7B · SynLogic (weak alignment)accuracy (%): 42.6 -> 10.2Hurt
replace ground-truth reward with incorrect reward (ablation)DAPO Llama-3.1-8B-Instruct · MATH500 (weak alignment)accuracy (%): 32.5 -> 26.4Hurt
replace ground-truth reward with majority-vote reward (ablation)DAPO Qwen2.5-7B · SynLogicaccuracy (%): 1.5 -> 2.8No clear effect
train with majority-vote reward (ablation)DAPO Qwen2.5-7B · MATH500accuracy (%): 40.8 -> 69.4Improved
TTRL on strongly aligned Qwen2.5-7BQwen2.5-7B · MATH500accuracy (%): 40.8 -> 62.1Improved
TTRL on weakly aligned taskQwen2.5-7B · SynLogicaccuracy (%): 1.5 -> 1.8No clear effect
one-shot RL instead of full dataset, strong alignment (ablation)DAPO Qwen2.5-7B · MATH500accuracy (%): 71.0 -> 65.2 (base 40.8)Mixed
one-shot RL on weakly aligned logic tasks (ablation)DAPO Qwen2.5-7B · SynLogicaccuracy (%): 1.5 -> 0.8Hurt
variance-based sample selection instead of random single sample (ablation)one-shot RL Qwen2.5-7B · MATH500accuracy (%): selected 65.2 vs random 58.7 and 63.0No clear effect
negative-sample-only training (NSR) instead of full DAPO, strong alignment (ablation)DAPO Qwen2.5-7B · MATH500accuracy (%): 71.0 -> 68.7No clear effect
negative-sample-only training (NSR) in weak alignment (ablation)DAPO Qwen2.5-7B · SynLogicaccuracy (%): 42.6 -> 1.5Hurt

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