RepBench: Compiling Benchmarks into Capability Representations for Large Language Models

2026/07/30 by Yanshi Li, Xueru Bai, Shuman Liu +1
Computer Science · #cs.CL #cs.AI

paper · pdf

22 pages, 8 figures, with appendices. Yanshi Li and Xueru Bai contributed equally

arxiv created 2026/07/30 · arxiv updated 2026/07/31

Abstract

Representation engineering reads and steers capability directions in large language models, yet methods are typically evaluated on paper-specific synthetic data. The resulting measurements are difficult to compare or reproduce and may reflect surface patterns rather than capabilities. We present RepBench, a benchmark-grounded data layer for capability-aligned representation probing. Crawling 13,427 benchmark papers yields a taxonomy of 182 capability clusters in 13 families; harvesting 353 public benchmark datasets yields 46,149 audited probe texts covering 94 capabilities, each supported by at least two independent benchmarks. This multi-benchmark design reduces dependence on any single source: raw per-text vectors exhibit no natural cluster granularity, whereas benchmark-pooled capability vectors show an interior clustering optimum at a small number of clusters on all 12 evaluated models, with low agreement to the human taxonomy. Under cross-benchmark transfer evaluation across twelve models completed by all four readouts, difference-in-means attains the highest model-level mean on ten models, while logistic regression wins the most capability-model cells. This disagreement shows that the readout method and aggregation criterion are meaningful evaluation dimensions. The pipeline, corpus, and evaluation code are released as a reusable closed-loop workflow.

Citations

• Verbalizable Representations Form a Global Workspace in Language Models
• Convergent Evolution: How Different Language Models Learn Similar Number Representations
• The Impact of Off-Policy Training Data on Probe Generalisation
• RepreGuard: Detecting LLM-Generated Text by Revealing Hidden Representation Patterns
• Is This Just Fantasy? Language Model Representations Reflect Human Judgments of Event Plausibility
• Activation Reward Models for Few-Shot Model Alignment
• Beyond Demonstrations: Dynamic Vector Construction from Latent Representations
• Calibrating Verbal Uncertainty as a Linear Feature to Reduce Hallucinations
• Toward universal steering and monitoring of AI models
• AxBench: Steering LLMs? Even Simple Baselines Outperform Sparse Autoencoders
• Exploring the LLM Journey from Cognition to Expression with Linear Representations
• Steering Llama 2 via Contrastive Activation Addition
• Steering Language Models With Activation Engineering

Related

• A Survey on Large Language Model Benchmarks
• STEM: Efficient Relative Capability Evaluation of LLMs through Structured Transition Samples
• SCAN: Structured Capability Assessment and Navigation for LLMs
• Mapping Overlaps in Benchmarks through Perplexity in the Wild
• Benchmark Profiling: Mechanistic Diagnosis of LLM Benchmarks