NoisyCoconut: Counterfactual Consensus via Latent Space Reasoning

arXiv cs.LG Papers
inference-time reliability latent-space llm abstention counterfactual-reasoning noisyCoconut

Summary

This paper introduces NoisyCoconut, an inference-time method that improves LLM reliability by injecting noise into latent trajectories to generate diverse reasoning paths. The approach enables models to abstain when uncertain, significantly reducing error rates in mathematical reasoning tasks without requiring retraining.

arXiv:2605.08221v1 Announce Type: new Abstract: This paper presents NoisyCoconut, a novel inference-time method that enhances large language model (LLM) reliability by manipulating internal representations. Unlike fine-tuning methods that require extensive retraining, NoisyCoconut operates directly on model representations during inference and requires no retraining. Rather than training models to reason in latent space, we inject controlled noise into latent trajectories to generate diverse reasoning paths. Agreement among these paths provides a confidence signal, enabling models to abstain when uncertain. We demonstrate that this approach achieves effective coverage-accuracy tradeoffs across multiple reasoning benchmarks without requiring access to training data or modification of model parameters. This approach provides a practical pathway to improving the reliability of LLM outputs while maintaining compatibility with existing models. Our experiments show that unanimous agreement among noise-perturbed paths reduces error rates from 40-70% to below 15%, enabling models to exceed 95% accuracy on mathematical reasoning tasks through selective abstention.
Original Article
View Cached Full Text

Cached at: 05/12/26, 07:08 AM 

# NoisyCoconut: Counterfactual Consensus via Latent Space Reasoning
Source: [https://arxiv.org/abs/2605.08221](https://arxiv.org/abs/2605.08221)
[View PDF](https://arxiv.org/pdf/2605.08221)

> Abstract:This paper presents NoisyCoconut, a novel inference\-time method that enhances large language model \(LLM\) reliability by manipulating internal representations\. Unlike fine\-tuning methods that require extensive retraining, NoisyCoconut operates directly on model representations during inference and requires no retraining\. Rather than training models to reason in latent space, we inject controlled noise into latent trajectories to generate diverse reasoning paths\. Agreement among these paths provides a confidence signal, enabling models to abstain when uncertain\. We demonstrate that this approach achieves effective coverage\-accuracy tradeoffs across multiple reasoning benchmarks without requiring access to training data or modification of model parameters\. This approach provides a practical pathway to improving the reliability of LLM outputs while maintaining compatibility with existing models\. Our experiments show that unanimous agreement among noise\-perturbed paths reduces error rates from 40\-70% to below 15%, enabling models to exceed 95% accuracy on mathematical reasoning tasks through selective abstention\.

## Submission history

From: Michael Jerge \[[view email](https://arxiv.org/show-email/c5dad695/2605.08221)\] **\[v1\]**Wed, 6 May 2026 13:58:55 UTC \(212 KB\)

Similar Articles