mlsanityData to 7 Oct 2026

Can LLMs Reconcile Knowledge Conflicts in Counterfactual Reasoning

arXiv 2506.15732 ↗First posted 15 Jun 2025cs.AI cs.LGcited by 7

On the map: LLM reasoning → Reasoning benchmarks and evaluation

Abstract (arXiv)

Large Language Models have been shown to contain extensive world knowledge in their parameters, enabling impressive performance on many knowledge intensive tasks. However, when deployed in novel settings, LLMs often encounter situations where they must integrate parametric knowledge with new or unfamiliar information. In this work, we explore whether LLMs can combine knowledge in-context with their parametric knowledge through the lens of counterfactual reasoning. Through synthetic and real experiments in multi-hop reasoning problems, we show that LLMs generally struggle with counterfactual reasoning, often resorting to exclusively using their parametric knowledge. Moreover, we show that simple post-hoc finetuning can struggle to instill counterfactual reasoning ability -- often leading to degradation in stored parametric knowledge. Ultimately, our work reveals important limitations of current LLM's abilities to re-purpose parametric knowledge in novel settings.

Key facts. Cited by 7 papers in the map (data through 7 Oct 2026). Placed in Reasoning benchmarks and evaluation, one of 17 directions in LLM reasoning. Builds on 29 of its 31 references that are in the map. First posted on arXiv 15 Jun 2025.

Builds on

29 of 31 references.

Closest ancestors · 1

Same direction, newest first

  1. 2024Grokked Transformers are Implicit Reasoners: A Mechanistic Journey to the Edge of Generalization23 May

Other references · 28

  1. 2022Training language models to follow instructions with human feedbackcited by 15,712
  2. 2019RoBERTa: A Robustly Optimized BERT Pretraining Approachcited by 13,266
  3. 2020Retrieval-Augmented Generation for Knowledge-Intensive NLP Taskscited by 9,160
  4. 2020Dense Passage Retrieval for Open-Domain Question Answeringcited by 3,779
  5. 2018HotpotQA: A Dataset for Diverse, Explainable Multi-hop Question Answeringcited by 3,686
  6. 2022Training a Helpful and Harmless Assistant with Reinforcement Learning from Human Feedbackcited by 3,634
  7. 2020Learning to summarize from human feedbackcited by 2,562
  8. 2020REALM: Retrieval-Augmented Language Model Pre-Trainingcited by 2,126
  9. 2019Language Models as Knowledge Bases?cited by 1,922
  10. 2020Leveraging Passage Retrieval with Generative Models for Open Domain Question Answeringcited by 1,203
  11. 2020How Much Knowledge Can You Pack Into the Parameters of a Language Model?cited by 713
  12. 2022Improving alignment of dialogue agents via targeted human judgementscited by 527
16 more
  1. 2023Causal Reasoning and Large Language Models: Opening a New Frontier for Causalitycited by 348
  2. 2024Do Large Language Models Latently Perform Multi-Hop Reasoning?cited by 165
  3. 2023Causal Parrots: Large Language Models May Talk Causality But Are Not Causalcited by 146
  4. 2023Physics of Language Models: Part 3.2, Knowledge Manipulationcited by 142
  5. 2023CLadder: Assessing Causal Reasoning in Language Modelscited by 102
  6. 2024Tug-of-War Between Knowledge: Exploring and Resolving Knowledge Conflicts in Retrieval-Augmented Language Modelscited by 93
  7. 2024Hopping Too Late: Exploring the Limitations of Large Language Models on Multi-Hop Queriescited by 78
  8. 2024Are LLMs Capable of Data-based Statistical and Causal Reasoning? Benchmarking Advanced Quantitative Reasoning with Datacited by 61
  9. 2023Faithful Explanations of Black-box NLP Models Using LLM-generated Counterfactualscited by 44
  10. 2024AdaCAD: Adaptively Decoding to Balance Conflicts between Contextual and Parametric Knowledgecited by 35
  11. 2024Discerning and Resolving Knowledge Conflicts through Adaptive Decoding with Contextual Information-Entropy Constraintcited by 27
  12. 2025Taming Knowledge Conflicts in Language Modelscited by 27
  13. 2024Do Large Language Models Perform Latent Multi-Hop Reasoning without Exploiting Shortcuts?cited by 26
  14. 2024Failure Modes of LLMs for Causal Reasoning on Narrativescited by 12
  15. 2025Grokking in the Wild: Data Augmentation for Real-World Multi-Hop Reasoning with Transformerscited by 12
  16. 2024Context-Parametric Inversion: Why Instruction Finetuning Can Worsen Context Reliancecited by 9

Built on by

Cited by year (2026 to 7 Oct)

Citing papers per year: 2025: 3; 2026: 4.

7 papers.

  1. 2025Executable Counterfactuals: Improving LLMs' Causal Reasoning Through Codecited by 10
  2. 2025A Unified Definition of Hallucination: It's The World Model, Stupid!cited by 6
  3. 2025MetaCrit: A Critical Thinking Framework for Self-Regulated LLM Reasoningcited by 1
  4. 2026Code over Words: Overcoming Semantic Inertia via Code-Grounded Reasoningcited by 1
  5. 2026Towards Generalizable Reasoning: Group Causal Counterfactual Policy Optimization for LLM Reasoningcited by 1
  6. 2026The Illusion of textitWhat If: Evaluating the Breakdown of Counterfactual Reasoning in LLMscited by 0
  7. 2026Who Warmed the Archives? LLMs Overestimate Historical Warmthcited by 0

Similar, not cited

Close in wording, no citation either way

  1. 2025Flying Pigs, FaR and Beyond: Evaluating LLM Reasoning in Counterfactual Worlds
  2. 2025On the Eligibility of LLMs for Counterfactual Reasoning: A Decompositional Study
  3. 2025CounterBench: Evaluating and Improving Counterfactual Reasoning in Large Language Models
  4. 2026Counterfactual Benchmarking and Training for Factuality Consistency and Order-Robust Grounded Reasoning in LLMs over Heterogeneous Knowledge
  5. 2023CLOMO: Counterfactual Logical Modification with Large Language Models
  6. 2026Exploring Knowledge Conflicts for Faithful LLM Reasoning: Benchmark and Method
  7. 2025Fast, Slow, and Tool-augmented Thinking for LLMs: A Review
  8. 2024Hierarchical Deconstruction of LLM Reasoning: A Graph-Based Framework for Analyzing Knowledge Utilization