# RepBench: Compiling Benchmarks into Capability Representations for Large Language Models

arXiv: https://arxiv.org/abs/2607.28008 · HTML: https://mlsanity.com/paper/2607.28008/

## Key facts

- No citing papers in the map yet (data through 14 Oct 2026).
- Placed in Knowledge and representations inside LLMs, one of 13 directions in Mechanistic interpretability.
- Builds on 26 of its 33 references that are in the map.
- First posted on arXiv 30 Jul 2026.

## Placed in

- Mechanistic interpretability / [Knowledge and representations inside LLMs](https://mlsanity.com/mechinterp/knowledge-and-representations-inside-llms/)

## Builds on (most cited first, 26 of 26)

- [TruthfulQA: Measuring How Models Mimic Human Falsehoods](https://mlsanity.com/paper/2109.07958/) (2021)
- [Steering Llama 2 via Contrastive Activation Addition](https://mlsanity.com/paper/2312.06681/) (2023)
- [Steering Language Models With Activation Engineering](https://mlsanity.com/paper/2308.10248/) (2023)
- [AxBench: Steering LLMs? Even Simple Baselines Outperform Sparse Autoencoders](https://mlsanity.com/paper/2501.17148/) (2025)
- [Contrastive Code Representation Learning](https://mlsanity.com/paper/2007.04973/) (2020)
- [Verbalizable Representations Form a Global Workspace in Language Models](https://mlsanity.com/paper/2607.15495/) (2026)
- [Calibrating Verbal Uncertainty as a Linear Feature to Reduce Hallucinations](https://mlsanity.com/paper/2503.14477/) (2025)
- [Toward universal steering and monitoring of AI models](https://mlsanity.com/paper/2502.03708/) (2025)
- [RepreGuard: Detecting LLM-Generated Text by Revealing Hidden Representation Patterns](https://mlsanity.com/paper/2508.13152/) (2025)
- [Spherical Steering: Geometry-Aware Activation Rotation for Language Models](https://mlsanity.com/paper/2602.08169/) (2026)
- [The Impact of Off-Policy Training Data on Probe Generalisation](https://mlsanity.com/paper/2511.17408/) (2025)
- [Why Steering Works: Toward a Unified View of Language Model Parameter Dynamics](https://mlsanity.com/paper/2602.02343/) (2026)
- [Rethinking LLM-as-a-Judge: Representation-as-a-Judge with Small Language Models via Semantic Capacity Asymmetry](https://mlsanity.com/paper/2601.22588/) (2026)
- [COLD-Steer: Steering Large Language Models via In-Context One-step Learning Dynamics](https://mlsanity.com/paper/2603.06495/) (2026)
- [Steer2Adapt: Dynamically Composing Steering Vectors Elicits Efficient Adaptation of LLMs](https://mlsanity.com/paper/2602.07276/) (2026)
- [What Drives Representation Steering? A Mechanistic Case Study on Steering Refusal](https://mlsanity.com/paper/2604.08524/) (2026)
- [Activation Reward Models for Few-Shot Model Alignment](https://mlsanity.com/paper/2507.01368/) (2025)
- [Convergent Evolution: How Different Language Models Learn Similar Number Representations](https://mlsanity.com/paper/2604.20817/) (2026)
- [Exploring the LLM Journey from Cognition to Expression with Linear Representations](https://mlsanity.com/paper/2405.16964/) (2024)
- [Beyond Demonstrations: Dynamic Vector Construction from Latent Representations](https://mlsanity.com/paper/2505.20318/) (2025)
- [Towards Intrinsic Interpretability of Large Language Models:A Survey of Design Principles and Architectures](https://mlsanity.com/paper/2604.16042/) (2026)
- [Is This Just Fantasy? Language Model Representations Reflect Human Judgments of Event Plausibility](https://mlsanity.com/paper/2507.12553/) (2025)
- [How Controllable Are Large Language Models? A Unified Evaluation across Behavioral Granularities](https://mlsanity.com/paper/2603.02578/) (2026)
- [Segment-Level Coherence for Robust Harmful Intent Probing in LLMs](https://mlsanity.com/paper/2604.14865/) (2026)
- [Understanding and Enforcing Weight Disentanglement in Task Arithmetic](https://mlsanity.com/paper/2604.17078/) (2026)
- [Decomposing Theory of Mind: How Emotional Processing Mediates ToM Abilities in LLMs](https://mlsanity.com/paper/2511.15895/) (2025)

## Abstract (arXiv)

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.