CLUE: Non-parametric Verification from Experience via Hidden-State Clustering

2025/10/02 by Zhenwen Liang, Liang, Zhenwen, Ruosen Li +13
Computer Science · Decision Sciences · #Computation and Language (cs.CL) #Data Visualization and Analytics #FOS: Computer and information sciences #Simulation Techniques and Applications #Time Series Analysis and Forecasting

paper · pdf · doi:10.48550/arxiv.2510.01591

openalex publication_date 2025/10/02 · openalex created_date 2025/10/10 · openalex updated_date 2026/07/28

Abstract

Assessing the quality of Large Language Model (LLM) outputs presents a critical challenge. Previous methods either rely on text-level information (e.g., reward models, majority voting), which can overfit to superficial cues, or on calibrated confidence from token probabilities, which would fail on less-calibrated models. Yet both of these signals are, in fact, partial projections of a richer source of information: the model's internal hidden states. Early layers, closer to token embeddings, preserve semantic and lexical features that underpin text-based judgments, while later layers increasingly align with output logits, embedding confidence-related information. This paper explores hidden states directly as a unified foundation for verification. We show that the correctness of a solution is encoded as a geometrically separable signature within the trajectory of hidden activations. To validate this, we present Clue (Clustering and Experience-based Verification), a deliberately minimalist, non-parametric verifier. With no trainable parameters, CLUE only summarizes each reasoning trace by an hidden state delta and classifies correctness via nearest-centroid distance to ``success'' and ``failure'' clusters formed from past experience. The simplicity of this method highlights the strength of the underlying signal. Empirically, CLUE consistently outperforms LLM-as-a-judge baselines and matches or exceeds modern confidence-based methods in reranking candidates, improving both top-1 and majority-vote accuracy across AIME 24/25 and GPQA. As a highlight, on AIME 24 with a 1.5B model, CLUE boosts accuracy from 56.7% (majority@64) to 70.0% (top-maj@16).

Citations

• Deep Think with Confidence
• Momentum Point-Perplexity Mechanics in Large Language Models
• Persona Vectors: Monitoring and Controlling Character Traits in Language Models
• ProRL: Prolonged Reinforcement Learning Expands Reasoning Boundaries in Large Language Models
• General-Reasoner: Advancing LLM Reasoning Across All Domains
• Mining Intrinsic Rewards from LLM Hidden States for Efficient Best-of-N Sampling
• Qwen3 Technical Report
• Efficient Reasoning with Hidden Thinking
• DeepSeek-R1 incentivizes reasoning in LLMs through reinforcement learning
• Multiple Choice Questions: Reasoning Makes Large Language Models (LLMs) More Self-Confident Even When They Are Wrong
• Reasoning Beyond Words ? Exploring framework for hidden state reasoning
• GPT-4o System Card
• Semantic Self-Consistency: Enhancing Language Model Reasoning via Semantic Weighting
• Scaling LLM Test-Time Compute Optimally can be More Effective than Scaling Model Parameters
• V-STaR: Training Verifiers for Self-Taught Reasoners
• GPQA: A Graduate-Level Google-Proof Q&A Benchmark
• A Survey of Confidence Estimation and Calibration in Large Language Models
• Judging LLM-as-a-Judge with MT-Bench and Chatbot Arena
• Let's Verify Step by Step
• Generating with Confidence: Uncertainty Quantification for Black-box Large Language Models
• Tree of Thoughts: Deliberate Problem Solving with Large Language Models
• Self-Refine: Iterative Refinement with Self-Feedback
• Eliciting Latent Predictions from Transformers with the Tuned Lens
• Affective Coherence Monitoring for Transformer-Based Language Models
• Improving alignment of dialogue agents via targeted human judgements
• Language Models (Mostly) Know What They Know
• Self-Consistency Improves Chain of Thought Reasoning in Language Models
• Training language models to follow instructions with human feedback
• Training Verifiers to Solve Math Word Problems
• Measuring Mathematical Problem Solving With the MATH Dataset

Cited by

• Evolving Language Models without Labels: Majority Drives Selection, Novelty Promotes Variation

Related

• Inside-Out: Hidden Factual Knowledge in LLMs
• LENS: Learning Ensemble Confidence from Neural States for Multi-LLM Answer Integration
• Just Ask for Calibration: Strategies for Eliciting Calibrated Confidence Scores from Language Models Fine-Tuned with Human Feedback
• Verifying Large Language Models' Reasoning Paths via Correlation Matrix Rank
• ConfTuner: Training Large Language Models to Express Their Confidence Verbally