mlsanityData to 14 Oct 2026

Causal Concept Graphs in LLM Latent Space for Stepwise Reasoning

arXiv 2603.10377 ↗First posted 11 Mar 2026cs.LG cs.AI stat.MEcited by 0

On the map: LLM reasoning → Causal reasoning in LLMs

also in 1 more
Abstract (arXiv)

Sparse autoencoders can localize where concepts live in language models, but not how they interact during multi-step reasoning. We propose Causal Concept Graphs (CCG): a directed acyclic graph over sparse, interpretable latent features, where edges capture learned causal dependencies between concepts. We combine task-conditioned sparse autoencoders for concept discovery with DAGMA-style differentiable structure learning for graph recovery and introduce the Causal Fidelity Score (CFS) to evaluate whether graph-guided interventions induce larger downstream effects than random ones. On ARC-Challenge, StrategyQA, and LogiQA with GPT-2 Medium, across five seeds (n=15 paired runs), CCG achieves CFS=5.654±0.625, outperforming ROME-style tracing (3.382±0.233), SAE-only ranking (2.479±0.196), and a random baseline (1.032±0.034), with p<0.0001 after Bonferroni correction. Learned graphs are sparse (5-6% edge density), domain-specific, and stable across seeds.

Key facts. No citing papers in the map yet (data through 14 Oct 2026). Placed in Causal reasoning in LLMs, one of 17 directions in LLM reasoning. Builds on 21 of its 32 references that are in the map. First posted on arXiv 11 Mar 2026.

Builds on

21 of 32 references.

References · 21

  1. 2014Adam: A Method for Stochastic Optimizationcited by 37,240
  2. 2022Chain-of-Thought Prompting Elicits Reasoning in Large Language Modelscited by 12,459
  3. 2018Think you have Solved Question Answering? Try ARC, the AI2 Reasoning Challengecited by 4,886
  4. 2022Large Language Models are Zero-Shot Reasonerscited by 4,374
  5. 2022Locating and Editing Factual Associations in GPTcited by 2,691
  6. 2023Sparse Autoencoders Find Highly Interpretable Features in Language Modelscited by 1,420
  7. 2022Interpretability in the Wild: a Circuit for Indirect Object Identification in GPT-2 smallcited by 1,165
  8. 2020Concept Bottleneck Modelscited by 1,032
  9. 2022Mass-Editing Memory in a Transformercited by 1,001
  10. 2022Toy Models of Superpositioncited by 971
  11. 2022In-context Learning and Induction Headscited by 964
  12. 2023Progress measures for grokking via mechanistic interpretabilitycited by 839
9 more
  1. 2021Did Aristotle Use a Laptop? A Question Answering Benchmark with Implicit Reasoning Strategiescited by 743
  2. 2024Scaling and evaluating sparse autoencoderscited by 553
  3. 2021Causal Abstractions of Neural Networkscited by 422
  4. 2022Post-hoc Concept Bottleneck Modelscited by 293
  5. 2023Linearity of Relation Decoding in Transformer Language Modelscited by 200
  6. 2023Does Circuit Analysis Interpretability Scale? Evidence from Multiple Choice Capabilities in Chinchillacited by 149
  7. 2022DAGMA: Learning DAGs via M-matrices and a Log-Determinant Acyclicity Characterizationcited by 140
  8. 2022Toward Transparent AI: A Survey on Interpreting the Inner Structures of Deep Neural Networkscited by 129
  9. 2023Interpretability at Scale: Identifying Causal Mechanisms in Alpacacited by 103

Built on by

Not cited yet.

Similar, not cited

Close in wording, no citation either way

  1. 2024Causal Graph Discovery with Retrieval-Augmented Generation based Large Language Models
  2. 2024CausalGraph2LLM: Evaluating LLMs for Causal Queries
  3. 2026Domain-Filtered Knowledge Graphs from Sparse Autoencoder Features
  4. 2026Causal-Audit: Explicit and Auditable Graph-based Reasoning via Target-Aware Causal Chain Construction
  5. 2026When Trees Are Not Enough: Learning Mixed-Topology Feature Graphs with Adaptive Graph Sparse Autoencoders
  6. 2026LLM Explainability with Counterfactual Chains and Causal Graphs
  7. 2024Narrating Causal Graphs with Large Language Models
  8. 2025CAMA: Enhancing Mathematical Reasoning in Large Language Models with Causal Knowledge