mlsanityData to 14 Oct 2026

How Do Linear Probes Emerge? A Circuit-Tracing Framework with Concept-Targeted Attribution

arXiv 2608.27510 ↗First posted 27 Aug 2026cs.CL cs.LGcited by 0

On the map: Mechanistic interpretability → Circuits and mechanistic analysis

Abstract (arXiv)

Transcoder attribution graphs are usually trained to explain why a model assigns high probability to a particular next token. We introduce Concept-Targeted Attribution (CTA), which instead trains attribution graphs with respect to a linear probe direction. CTA therefore yields probe-specific circuits that explain why an internal concept representation arises in a prompt, independently of whether it is expressed in the generated token. Using Cross-Layer Transcoders, we show that these probe-targeted graphs contain predictive structure: graph-level features predict probe accuracy across four widely studied concept categories (ρ = 0.91, R2 = 0.84), while local features identify the sparse components driving per-prompt classification. This connects probe performance to interpretable circuit structure, allowing us to ask not only whether a probe works, but which internal computations make it work. Causal ablations further show that probe-targeted and logit-targeted graphs capture functionally distinct mechanisms. Removing probe-relevant features reduces internal concept scores while largely preserving generated tokens, whereas removing logit-relevant features changes the generated token in 92% to 100% of cases with near-zero effect on probe scores. CTA provides a framework for moving from behavioral probe accuracy to mechanistic explanations of probe performance, enabling more detailed audits of internal concept representations, including safety-critical ones. Our code is available at https://github.com/vedantpalit/concept-targeted-attribution

Key facts. No citing papers in the map yet (data through 14 Oct 2026). Placed in Circuits and mechanistic analysis, one of 13 directions in Mechanistic interpretability. Builds on 20 of its 32 references that are in the map. First posted on arXiv 27 Aug 2026.

Builds on

20 of 32 references.

Closest ancestors · 5

Same direction, newest first

  1. 2026All Circuits Lead to Rome: Rethinking Functional Anisotropy in Circuit and Sheaf Discovery for LLMs12 May
  2. 2024Transcoders Find Interpretable LLM Feature Circuits17 Jun
  3. 2024Mechanistic Interpretability for AI Safety -- A Review22 Apr
  4. 2023Towards Automated Circuit Discovery for Mechanistic Interpretability28 Apr
  5. 2022Interpretability in the Wild: a Circuit for Indirect Object Identification in GPT-2 small1 Nov

Other references · 15

  1. 2018HotpotQA: A Dataset for Diverse, Explainable Multi-hop Question Answeringcited by 3,754
  2. 2017TriviaQA: A Large Scale Distantly Supervised Challenge Dataset for Reading Comprehensioncited by 2,955
  3. 2022Locating and Editing Factual Associations in GPTcited by 2,691
  4. 2023Sparse Autoencoders Find Highly Interpretable Features in Language Modelscited by 1,420
  5. 2016Understanding intermediate layers using linear classifier probescited by 1,353
  6. 2023Representation Engineering: A Top-Down Approach to AI Transparencycited by 1,336
  7. 2017Interpretability Beyond Feature Attribution: Quantitative Testing with Concept Activation Vectors (TCAV)cited by 1,319
  8. 2022Discovering Latent Knowledge in Language Models Without Supervisioncited by 767
  9. 2023The Geometry of Truth: Emergent Linear Structure in Large Language Model Representations of True/False Datasetscited by 712
  10. 2020Causal Mediation Analysis for Interpreting Neural NLP: The Case of Gender Biascited by 191
  11. 2026Scaling Monosemanticity: Extracting Interpretable Features from Claude 3 Sonnetcited by 191
  12. 2025The Reasoning-Memorization Interplay in Language Models Is Mediated by a Single Directioncited by 21
3 more
  1. 2025Verifying Chain-of-Thought Reasoning via Its Computational Graphcited by 20
  2. 2025Tracing Multilingual Representations in LLMs with Cross-Layer Transcoderscited by 8
  3. 2026CLT-Forge: A Scalable Library for Cross-Layer Transcoders and Attribution Graphscited by 2

Built on by

Not cited yet.

Similar, not cited

Close in wording, no citation either way

  1. 2025Automated Attribution Graph Interpretation via Probe Prompting
  2. 2026Circuit Condensation: Post-Training that Concentrates a Behavior's Causal Circuit
  3. 2026Many Circuits, One Mechanism: Input Variation and Evaluation Granularity in Circuit Discovery
  4. 2025Attribution-Guided Pruning for Insight and Control: Circuit Discovery and Targeted Correction in Small-scale LLMs
  5. 2026Slaying the Hydra: Interaction-Aware Circuit Discovery in Language Models
  6. 2026Finding Interpretable Prompt-Specific Circuits in Language Models
  7. 2023Attribution Patching Outperforms Automated Circuit Discovery
  8. 2026Every Component Is a Lookup: One Linear Graph for Interaction, Composition and Attribution