---
title: "Can We Break LLMs Out of Self-Loops? Fine-Grained Reasoning Control with Activation Steering"
source: "arXiv AI/CL/LG（论文）"
source_tier: "t2"
url: "https://arxiv.org/abs/2607.18100v1"
permalink: "https://feed.astralor.com/items/itm_094dafceb27cedaa555fd4f4"
published: "2026-07-20T16:01:57.000Z"
score: 79
channel: "ai"
tags: ["LLM Inference","SOPHIA","Reasoning Control","Activation Steering","Hidden-state Intervention"]
---

# Can We Break LLMs Out of Self-Loops? Fine-Grained Reasoning Control with Activation Steering

## AI commentary

> This paper offers a fresh perspective on solving the 'reasoning black hole' problem in LLMs, crucial for researchers in reasoning controllability and AI safety. SOPHIA's offline construction of steering vectors for online intervention is practical and shows clear token efficiency gains. I particularly appreciate the framing of reasoning traces as state sequences, which opens up new avenues for future work.

## At a glance

This paper analyzes self-loop failure trajectories in LLM reasoning, and proposes SOPHIA, a method that treats reasoning traces as sequences of latent states. By classifying prefixes into states, recording transitions, and constructing a bank of steering vectors indexed by state pairs, a controller can detect self-loops online and apply targeted interventions. Experiments show that SOPHIA reliably prevents self-loops, improving end-task accuracy and token efficiency, demonstrating fine-grained controllability over reasoning processes.

## Key points

- Identifies self-loop failures in LLM reasoning trajectories that exhaust token budgets without progress
- Proposes SOPHIA, using latent state classification and state-pair-indexed steering vectors for inference-time intervention
- Empirically demonstrates that the method reliably reduces self-loops, improving reasoning quality and token efficiency

## Full text

Extended reasoning has become standard for frontier Large Language Models (LLMs), yet the trajectories these models produce remain largely uncontrollable. Existing methods for shaping how a model reasons are prompt based approaches and operate at the input level, offering no fine-grained control over the reasoning process itself. Related work analyzes and discovers latent transition dynamics in the reasoning traces from Large Language Models. Building on this, we statistically characterize these states, and show that failure trajectories get stuck in self-loops, exhausting the token budget without progress toward the final answer. To intervene on these failures, We propose SOPHIA: Steering Of reasoning Processes via Hidden-state Intervention and Activations. We treat each reasoning trace as a sequence of latent states rather than an unstructured texts, and investigate whether inference time interventions can provide fine-grained control over the self-looping reasoning process. We classify every prefix to a latent state, record step level transitions, and use them to construct a bank of steering vectors indexed by state pairs. At inference time, a controller infers the current state and, given a target state, retrieves the corresponding vector and can also detect self-loops online from the transition structure to prevent the model from sinking into a reasoning black hole. Through extensive experiments, our method reliably intervenes on self-loop failures, with steering vectors that generalize to different state pairs. End task accuracy and token efficiency indicate that fine-grained controllability results in better reasoning quality.

---

Verify against the original before citing; curated by Astral Feed: https://feed.astralor.com/items/itm_094dafceb27cedaa555fd4f4