# Steering LRMs Beyond Output Degradation
_A new probe-based method, FPCG, distinguishes prediction from detection features to enable precise large reasoning models steering with minimal output quality degradation._
**Published:** 2026-06-10
**Source:** https://www.startuphub.ai/ai-news/ai-research/2026/steering-lrms-beyond-output-degradation
---
Deployed large [reasoning](/ai-news/ai-research/2026/llm-reasoning-fix-lpsr) models (LRMs) frequently exhibit unpredictable behaviors, a challenge that test-time steering methods have attempted to address. However, existing approaches often degrade output quality by relying on internal features that detect already generated text, rather than predicting future outcomes.

LRM Output DegradationDriver

deployed large reasoning models often exhibit unpredictable behaviorsFrom the articleFPCG enables precise large reasoning models steering with remarkably little degradation in output quality, a significant improvement over previous methods.problemExisting Steering MethodsDriverrely on internal features detecting already generated textFrom the article 5 mentionsPrior steering techniques inadvertently focused on features that signal existing behavior, which proved to be poor indicators of future actions.flawDetection vs PredictionContextdistinguishing features that signal existing vs future behaviorFrom the article 4 mentionsThe core innovation presented by Kortukov, Komorowski, and colleagues in their arXiv preprint lies in identifying a critical distinction between detection and prediction features within LRM hidden states.solutionActivation ProbesCoreFrom the article 5 mentionsThis paper introduces activation probes trained to forecast future behavior likelihoods from intermediate reasoning steps.Predicting Future BehaviorEffectprobes demonstrate significant accuracy from 64% to 91%From the article 5 mentionsHowever, existing approaches often degrade output quality by relying on internal features that detect already generated text, rather than predicting future outcomes.FPCG MethodCorefuture probe controlled generation enables precise steeringFrom the article 5 mentionsFurthermore, FPCG demonstrates efficacy in steering scenarios where activation steering methods fail, underscoring its robustness and broader applicability.achievesMinimal Quality DegradationOutcomeFrom the articleFPCG enables precise large reasoning models steering with remarkably little degradation in output quality, a significant improvement over previous methods.

## Unmasking Prediction Features for Control

The core innovation presented by Kortukov, Komorowski, and colleagues in their [arXiv](https://arxiv.org/abs/2606.11172v1) preprint lies in identifying a critical distinction between detection and prediction features within LRM hidden states. Prior steering techniques inadvertently focused on features that signal existing behavior, which proved to be poor indicators of future actions. This paper introduces activation probes trained to forecast future behavior likelihoods from intermediate reasoning steps. These probes demonstrate significant accuracy, ranging from 64% to 91%, in predicting the most probable behavior, thereby revealing a distinct set of internal prediction features.

## Future Probe Controlled Generation: Precision Steering

Building upon these newly identified prediction features, the authors propose [Future Probe Controlled Generation](/ai-news/ai-research/2026/llms-leap-from-knowledge-to-innovation) (FPCG). This novel text-level steering method enhances control by sampling multiple candidate sentences and selecting the optimal one based on a probe's prediction of future behavior likelihood. FPCG enables precise large reasoning models steering with remarkably little degradation in output quality, a significant improvement over previous methods. Furthermore, FPCG demonstrates efficacy in steering scenarios where activation steering methods fail, underscoring its robustness and broader applicability.
---
Original analysis from [startuphub.ai](https://www.startuphub.ai), the #1 AI startup directory.