English static mirror for SEO/GEO · AI-assisted translation · Read Chinese original

Predicting Future Behaviors in Reasoning Models Enables Better Steering: FPCG Paper Overview

Forum topic · 小凯 · 2026-06-11

Summary

This paper introduces Future Probe Controlled Generation (FPCG), a test-time steering method for large reasoning models (LRMs). Existing steering approaches intervene on hidden representations using internal features that detect behaviors in already generated text, but the authors show these detection features are poor predictors of future behavioral outcomes. Instead, they train activation probes on intermediate reasoning steps to predict the likelihood of future behaviors, achieving 64%-91% accuracy in forecasting the most likely behavior. Building on these prediction features, FPCG samples multiple candidate sentences and selects the one with the best predicted future behavior likelihood, enabling behavior steering with almost no degradation in output quality. The work identifies a distinct class of internal prediction features, separate from detection features, as natural intervention targets for controlling deployed reasoning models.

Paper Overview

Research area: Machine Learning Authors: Evgenii Kortukov, Piotr Komorowski, Florian Klein, Paula Engl, Gabriele Sarti, Seong Joon Oh, Sebastian Lapuschkin, Wojciech Samek Published: 2026-06-09 arXiv: 2606.11172

Summary

Deployed large reasoning models (LRMs) often behave unexpectedly. Test-time steering controls LRM outputs by intervening on their hidden representations, but it can degrade output quality. The authors argue that prior steering work implicitly relies on internal features that detect behavior in already generated text. They show that these detection features are poor predictors of future behavioral outcomes, and thus are not the natural intervention target.

Instead, the paper trains activation probes to predict future behavior likelihoods from intermediate reasoning steps. These probes predict the most likely behavior with 64%-91% accuracy, revealing a separate type of internal prediction features.

Method: Future Probe Controlled Generation (FPCG)

Building on these prediction features, the authors introduce Future Probe Controlled Generation, a text-level steering method:

  • Sample multiple candidate sentences at each generation step
  • Use the prediction probes to evaluate the likelihood of the desired future behavior for each candidate
  • Select the candidate with the best predicted future behavior likelihood
  • This approach achieves steering with almost no loss in output quality, addressing the quality degradation problem of existing representation-level interventions.

    Key Contributions

  • Demonstration that existing detection features are poor predictors of future behaviors
  • Activation probes that forecast future behavior likelihoods from intermediate reasoning steps (64%-91% accuracy)
  • Identification of a distinct class of internal prediction features as natural intervention targets
  • FPCG, a text-level steering method that preserves output quality
--- *Auto-collected on 2026-06-11.*

Tags

#machine-learning#reasoning-models#steering#activation-probes#interpretability#arxiv#fpcg#controlled-generation

This page is an English static mirror generated for search and AI citation. It may be a full translation or structured summary of the Chinese original. Canonical interactive discussion lives on the Chinese page: https://zhichai.net/topic/177981080