ThoughtProbe: Classifier-Guided LLM Thought Space Exploration via Probing Representations

2025/10/31 by Zijian Wang, Chang Xu, Wang, Zijian +1 · 1 citation
Computer Science · #Advanced Graph Neural Networks #Computation and Language (cs.CL) #Explainable Artificial Intelligence (XAI) #FOS: Computer and information sciences #Topic Modeling

paper · pdf · doi:10.48550/arxiv.2510.27355

openalex publication_date 2025/10/31 · openalex created_date 2025/11/05 · openalex updated_date 2026/07/28

Abstract

This paper introduces ThoughtProbe, a novel inference time framework that leverages the hidden reasoning features of Large Language Models (LLMs) to improve their reasoning performance. Unlike previous works that manipulate the hidden representations to steer LLM generation, we harness them as discriminative signals to guide the tree structured response space exploration. In each node expansion, a classifier serves as a scoring and ranking mechanism that efficiently allocates computational resources by prioritizing higher score candidates for continuation. After completing the tree expansion, we collect answers from all branches to form a candidate answer pool. We then propose a branch aggregation method that marginalizes over all supporting branches by aggregating their CoT scores, thereby identifying the optimal answer from the pool. Experimental results show that our framework's comprehensive exploration not only covers valid reasoning chains but also effectively identifies them, achieving significant improvements across multiple arithmetic reasoning benchmarks.

Citations

• Improving Reasoning Performance in Large Language Models via Representation Engineering
• Steering off Course: Reliability Challenges in Steering Language Models
• The Reasoning-Memorization Interplay in Language Models Is Mediated by a Single Direction
• Is Best-of-N the Best of Them? Coverage, Scaling, and Optimality in Inference-Time Alignment
• Unlocking General Long Chain-of-Thought Reasoning Capabilities of Large Language Models via Representation Engineering
• Rethinking Bradley-Terry Models in Preference-Based Reward Modeling: Foundations, Theory, and Alternatives
• Fast Best-of-N Decoding via Speculative Rejection
• Steering Knowledge Selection Behaviours in LLMs via SAE-Based Representation Engineering
• HaloScope: Harnessing Unlabeled LLM Generations for Hallucination Detection
• Programming Refusal with Conditional Activation Steering
• Physics of Language Models: Part 2.1, Grade-School Math and the Hidden Reasoning Process
• Advancing Process Verification for Large Language Models via Tree-Based Preference Learning
• SEED: Accelerating Reasoning Tree Construction via Scheduled Speculative Decoding
• A Language Model's Guide Through Latent Space
• Chain-of-Thought Reasoning Without Prompting
• INSIDE: LLMs' Internal States Retain the Power of Hallucination Detection
• Training Chain-of-Thought via Latent-Variable Inference
• The Linear Representation Hypothesis and the Geometry of Large Language Models
• Textbooks Are All You Need II: phi-1.5 technical report
• Large Language Models
• Inference-Time Intervention: Eliciting Truthful Answers from a Language Model
• Large Language Model Guided Tree-of-Thought
• Discovering Latent Knowledge in Language Models Without Supervision
• Large Language Models are Zero-Shot Reasoners
• Self-Consistency Improves Chain of Thought Reasoning in Language Models
• Are NLP Models really able to Solve Simple Math Word Problems?
• Attention Is All You Need
• Solving General Arithmetic Word Problems
• RANK ANALYSIS OF INCOMPLETE BLOCK DESIGNS
• Steerable Chatbots: Exploring Personalization Control Interfaces via LLM Activation Steering

Cited by

• CAST: Mitigating Object Hallucination in Large Vision-Language Models via Caption-Guided Visual Attention Steering

Related

• ThoughtProbe: Classifier-Guided Thought Space Exploration Leveraging LLM Intrinsic Reasoning
• Blind Spot Navigation in Large Language Model Reasoning with Thought Space Explorer
• Forest-of-Thought: Scaling Test-Time Compute for Enhancing LLM Reasoning
• Can Small Language Models Help Large Language Models Reason Better?: LM-Guided Chain-of-Thought
• SmartSwitch: Advancing LLM Reasoning by Overcoming Underthinking via Promoting Deeper Thought Exploration