AI Paper Atlas

Research map / Large language models / Reasoning and chain of thought

Patching Gaps In LLM Reasoning With Interventional Training

Matthew Y. R. Yang, Hao Bai, Ian Wu, Gene Yang, Amrith Setlur, Aviral Kumar

ICLR 2026 · Large language models › Reasoning and chain of thought · approach: RL with verifiable rewards for reasoning

Read the paperCheck your idea against 137,217 papers

AI-extracted summary. Written by AI from the paper's full text. Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.

Proposes interventional training to address RL stalling on hard problems where LLM policies never sample correct rollouts, by patching reasoning gaps (based only on the abstract).

Problem

RL training of LLMs stalls on hard problems because the policy rarely generates non-zero-reward rollouts, so it receives no learning signal; the abstract notes hard problems often remain unsolved due to repeated generation of incorrect reasoning steps.

Approach

Uses interventions on the reasoning process during training to inject correct steps and patch gaps in LLM reasoning so that hard problems yield non-zero-reward rollouts (details not stated in the abstract).

Measured effects reported in the paper

What was changedWhereMeasuredEffect
condition rollouts on a self-proposed intervention instead of the original erroneous step (ablation)Qwen3-4B-Instruct-2507 · Omni-MATH hard subset (334 problems)accuracy: 0.0713% -> 1.56%; coverage 29/334 -> 80/334Improved
clone the prefix before the intervention during SFT (ablation)InT SFT · 235 DeepScaleR problemsaccuracy: 2.87% -> 7.71%; coverage 162/235 -> 202/235Improved
also clone the correct suffix in SFT (ablation)InT SFT · 235 DeepScaleR problemsaccuracy: 7.71% -> 2.31%; coverage 202/235 -> 111/235Hurt
filter interventions to those yielding a correct rollout in 32 samples (ablation)InT SFT · 235 DeepScaleR problemsaccuracy: 5.06% -> 7.71%; coverage 196/235 -> 202/235Improved
give the reference solution when proposing interventions (ablation)intervention generation · Omni-MATH hard subset (334 problems)accuracy: 0.171% -> 1.56%Improved
use larger Qwen3-30B-A3B-Instruct to propose interventions (ablation)intervention generation · Omni-MATH hard subset (334 problems)accuracy: 1.56% -> 2.87%; coverage 80 -> 101Improved
use reasoning-tuned Qwen3-4B instead of instruct model to propose interventions (ablation)intervention generation · Omni-MATH hard subset (334 problems)coverage: coverage 80/334 -> 73/334; accuracy 1.56% -> 2.65%Mixed
combine hints with interventions (ablation)intervention rollouts · hint-guided pass@16=0 subsetcoverage: 18/176 -> 25/176Improved
add InT SFT before RLRL on Qwen3-4B-Instruct-2507 · IMO-AnswerBenchaccuracy: 23.46 -> 25.62Improved
SFT on reference solutions before RLRL on Qwen3-4B-Instruct-2507 · average of four benchmarksaverage score: 21.17 -> 20.76Hurt
SFT on self-reflection rewrites before RL instead of InTRL on Qwen3-4B-Instruct-2507 · AMO-Bench and IMO-AnswerBenchpass@8 / accuracy: AMO-Bench 36.72 vs 36.16; IMO-AnswerBench 15.53 vs 25.62Mixed
InT + RL versus standard RL on average scoreQwen3-4B-Instruct-2507 · average of four benchmarksaverage score: 28.26 -> 33.72Improved

Surprising findings

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