AI Paper Atlas

Research map / Reinforcement learning / RL from feedback

Enough is as good as a feast: A Comprehensive Analysis of How Reinforcement Learning Mitigates Task Conflicts in LLMs

Zixuan Ren, Jinliang Lu, Junhong Wu, Yang Zhao, Dai Dai, Hua Wu, Haifeng Wang, Chengqing Zong

ICLR 2026 · Reinforcement learning › RL from feedback · approach: RL for LLM reasoning · 1 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
train task models with GRPO instead of SFT before mergingLlama-3.1-8B, TIEs merge · Five-task averageavg accuracy after merge: 49.9 -> 67.1Improved
train task models with GRPO instead of SFT before mergingLlama-3.1-8B, Averaging · Five-task averageavg accuracy after merge: 47.9 -> 60.1Improved
use PPO or Reinforce++ instead of GRPO (ablation)RL merging · Math/Code/IF, TIEsavg accuracy after merge: GRPO 72.0, PPO 69.3, Reinforce++ 68.4No clear effect
use RL instead of SFT on smaller/larger base model (ablation)Llama-3.2-3B, Mistral-Small-24B · Math/Code/IF, TIEsavg accuracy after merge: 3B: 35.1 -> 46.7; 24B: 70.4 -> 80.4Improved
use on-policy data (RL, RFT) instead of fixed SFT data (ablation)training paradigm · Math update normparameter update norm: 6.5 -> 0.78 (RL), 2.36 (RFT)Improved
remove negative-sample advantages from RL (RL-Pos) (ablation)RL · Math/Code/IFsingle-task accuracy and merge degradation: RL-Pos below RL but above SFTHurt
add noise perturbation in random direction (ablation)SFT and RL models · performance landscapetask performanceNo clear effect
increase number of merged models (TIEs) (ablation)SFT vs RL · average performanceavg performance: SFT steeper decline than RLHurt
merge Puzzle-task models with TIEs (RL vs SFT)GRPO · Puzzleaccuracy: SFT 35.8 vs RL 64.6; RL still drops 24%Mixed

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