Reasoning Denoiser: Denoising Reasoning Traces for Hallucination Detection in Large Reasoning Models

Hugging Face Daily Papers Papers
reasoning hallucination-detection denoising large-reasoning-models chain-of-thought truthfulness

Summary

Introduces ReDe, a framework that denoises reasoning traces by filtering irrelevant and repetitive steps to improve hallucination detection in large reasoning models, achieving up to 87.32 AUROC on TruthfulQA.

Large reasoning models (LRMs) generate long reasoning traces before producing final answers. While these traces may contain useful signals for hallucination detection, harnessing them is non-trivial because long trajectories often include noisy steps that obscure the cues relevant to truthfulness assessment. In this paper, we identify two prevalent forms of reasoning noises, i.e., irrelevant steps and repetitive steps, and show that both substantially degrade hallucination detection performance. Existing confidence-based scores and naive embedding-based filtering fail to reliably separate noisy from informative steps. To address this challenge, we propose REDE, a novel learning framework for denoising reasoning traces for hallucination detection. Specifically, REDE leverages final-answer attention as an automatic supervision signal to shape the step-level representation space, yielding refined embeddings in which noisy steps can be reliably identified and filtered. REDE can be readily plugged into diverse hallucination detectors by operating on the filtered reasoning trajectory after removing noisy steps. Extensive experiments on multiple reasoning benchmarks show that REDE consistently improves detection performance over competitive baselines.
Original Article
View Cached Full Text

Cached at: 07/28/26, 10:24 AM

Paper page - Reasoning Denoiser: Denoising Reasoning Traces for Hallucination Detection in Large Reasoning Models

Source: https://huggingface.co/papers/2607.22098 👋 Hi Hugging Face community! We’re excited to share our new paper:

https://huggingface.co/papers/2607.22098#%F0%9F%A7%A0-reasoning-denoiser-denoising-reasoning-traces-for-hallucination-detection-in-large-reasoning-models🧠 Reasoning Denoiser: Denoising Reasoning Traces for Hallucination Detection in Large Reasoning Models

Large reasoning models generate long chains of thought, but are all reasoning steps useful for detecting hallucinations?

We find that reasoning traces frequently contain two types of noise:

🔸Irrelevant stepsthat contribute little to solving the problem 🔸Repetitive stepswhose information is already captured elsewhere

These noisy steps can obscure the signals needed to determine whether the final answer is correct.

To address this, we introduceReDe, a lightweight framework that:

✅ Uses final-answer attention as automatic supervision ✅ Learns representations that separate informative and noisy steps ✅ Filters reasoning traces without human step-level annotations ✅ Can be combined with probing-, uncertainty-, and verbalization-based detectors

📈 Across TruthfulQA, MATH, CodeElo, and MultiHopQA, ReDe consistently improves hallucination detection on Qwen3 and DeepSeek-R1 models. On TruthfulQA, it improves AUROC by up to18.69 percentage points, reaching87.32 AUROC.

Our main takeaway is simple:

For reliable reasoning models, we should not only analyze the reasoning trace—we should denoise it first.

📄 Paper:https://arxiv.org/abs/2607.22098

We would love to hear the community’s thoughts:What other downstream applications could benefit from filtering noisy reasoning steps?

Similar Articles

Reasoning Jury: Multi-Model Consensus for Evaluating Reasoning Traces

arXiv cs.AI

This paper introduces Reasoning Jury, a system that uses a jury of open-weight LLMs with a moderated consensus mechanism to evaluate long reasoning traces, significantly outperforming frontier models at identifying reasoning defects while costing a fraction of the price.