A Mechanistic Lens on Semantic Conflicts: Using Activation Patching to Understand LLM Behavior

• URL: http://arxiv.org/abs/2607.05587v1
• Date: Mon, 06 Jul 2026 19:35:04 GMT
• Title: A Mechanistic Lens on Semantic Conflicts: Using Activation Patching to Understand LLM Behavior
• Abstract summary: We present the first controlled, mechanistic study of large language models (LLMs) behavior under semantic conflicts.<n>We evaluate four open-weight LLMs on two tasks (output prediction and unit-test generation) using behavioral performance measures.<n>Our results show that semantic conflicts significantly reduce execution-grounded correctness in both tasks.
• Score: 7.506391522879222
• License: http://creativecommons.org/licenses/by-sa/4.0/
• Abstract: Large language models (LLMs) are increasingly used in software-engineering tasks processing executable code and non-executable semantic cues such as comments or identifiers. These two sources can conflict when semantic cues suggest different program behavior than the code itself. It remains unclear how such semantic conflicts affect LLM behavior and which source dominates their outputs. We present the first controlled, mechanistic study of LLM behavior under semantic conflicts. To this end, we construct 45 Python snippet triplets that isolate conflicts by varying either semantic cues or implementation while keeping token-aligned pairs for causal intervention. We evaluate four open-weight LLMs on two tasks (output prediction and unit-test generation) using behavioral performance measures and residual-stream activation patching to identify token-layer states that causally contribute to behavioral differences between aligned and conflicting inputs. Our results show that semantic conflicts significantly reduce execution-grounded correctness in both tasks and that all tested LLMs often follow misleading semantic cues. Residual-stream activation patching reveals a consistent pattern for final-output prediction: The changed cue/code region and a small set of intermediate tokens carry most of the recoverable causal signal before aggregation near the output readout. For unit-test generation, this pattern extends beyond the prompt, showing that conflict-related information is recoverable at generated sites before producing expected values. Overall, our findings show that semantic conflicts affect program comprehension and downstream tasks, with relevant information concentrated in a small number of causally active residual-stream states, and demonstrate a framework for mechanistically analyzing how LLMs integrate code-related information under controlled semantic variations.

Related papers

• PRIME: Evaluating Prompt Resolution Under Incompatible Instructions in LLMs [1.7243216387069673]
  Large language models (LLMs) often encounter conflicting prompts.<n>Current instruction following benchmarks assess those meta-instructions in isolation.<n>We introduce a framework textitPRIME to analyze behavior of LLMs when provided with conflicting instructions.
  arXiv  Detail & Related papers  (2026-06-21T12:30:23Z)
• From Early Encoding to Late Suppression: Interpreting LLMs on Character Counting Tasks [49.57538588967748]
  Large language models (LLMs) exhibit failures on elementary symbolic tasks such as character counting in a word, despite excelling on complex benchmarks.<n>We uncover a consistent phenomenon across modern architectures, including LLaMA, Qwen, and Gemma.<n>We show that symbolic reasoning failures in LLMs are not due to missing representations or insufficient scale, but arise from structured interference within the model's computation graph.
  arXiv  Detail & Related papers  (2026-04-01T11:40:12Z)
• On the Paradoxical Interference between Instruction-Following and Task Solving [50.75960598434753]
  Instruction following aims to align Large Language Models (LLMs) with human intent by specifying explicit constraints on how tasks should be performed.<n>We reveal a counterintuitive phenomenon: instruction following can paradoxically interfere with LLMs' task-solving capability.<n>We propose a metric, SUSTAINSCORE, to quantify the interference of instruction following with task solving.
  arXiv  Detail & Related papers  (2026-01-29T17:48:56Z)
• ConInstruct: Evaluating Large Language Models on Conflict Detection and Resolution in Instructions [26.40258251641021]
  This dataset is a benchmark for evaluating Large Language Models' ability to detect and resolve conflicts within user instructions.<n>Most proprietary LLMs exhibit strong conflict detection capabilities, whereas among open-source models, only DeepSeek-R1 demonstrates similarly strong performance.<n>Despite their strong conflict detection abilities, LLMs rarely explicitly notify users about the conflicts or request clarification when faced with conflicting constraints.
  arXiv  Detail & Related papers  (2025-11-18T10:49:37Z)
• Hybrid Fuzzing with LLM-Guided Input Mutation and Semantic Feedback [0.0]
  I present a hybrid fuzzing framework that integrates static and dynamic analysis with Large Language Model (LLM)-guided input mutation and semantic feedback.<n>Our method achieves faster time-to-first-bug, higher semantic diversity, and a competitive number of unique bugs compared to state-of-the-art fuzzers.
  arXiv  Detail & Related papers  (2025-11-06T02:38:24Z)
• InspectCoder: Dynamic Analysis-Enabled Self Repair through interactive LLM-Debugger Collaboration [71.18377595277018]
  Large Language Models (LLMs) frequently generate buggy code with complex logic errors that are challenging to diagnose.<n>We present InspectCoder, the first agentic program repair system that empowers LLMs to actively conduct dynamic analysis via interactive debugger control.
  arXiv  Detail & Related papers  (2025-10-21T06:26:29Z)
• DeLeaker: Dynamic Inference-Time Reweighting For Semantic Leakage Mitigation in Text-to-Image Models [55.30555646945055]
  Text-to-Image (T2I) models are vulnerable to semantic leakage.<n>We introduce DeLeaker, a lightweight approach that mitigates leakage by directly intervening on the model's attention maps.<n>SLIM is the first dataset dedicated to semantic leakage.
  arXiv  Detail & Related papers  (2025-10-16T17:39:21Z)
• Where Did It Go Wrong? Attributing Undesirable LLM Behaviors via Representation Gradient Tracing [12.835224376066769]
  Large Language Models (LLMs) have demonstrated remarkable capabilities, yet their deployment is frequently undermined by undesirable behaviors.<n>We introduce a novel and efficient framework that diagnoses a range of undesirable LLM behaviors by analyzing representation and its gradients.<n>We systematically evaluate our method for tasks that include tracking harmful content, detecting backdoor poisoning, and identifying knowledge contamination.
  arXiv  Detail & Related papers  (2025-09-26T12:07:47Z)
• LIRA: Inferring Segmentation in Large Multi-modal Models with Local Interleaved Region Assistance [54.683384204063934]
  Large multi-modal models (LMMs) struggle with inaccurate segmentation and hallucinated comprehension.<n>We propose LIRA, a framework that capitalizes on the complementary relationship between visual comprehension and segmentation.<n>LIRA achieves state-of-the-art performance in both segmentation and comprehension tasks.
  arXiv  Detail & Related papers  (2025-07-08T07:46:26Z)
• Interpreting and Steering LLMs with Mutual Information-based Explanations on Sparse Autoencoders [29.356200147371275]
  Large language models (LLMs) excel at handling human queries, but they can occasionally generate flawed or unexpected responses.<n>We propose using a fixed vocabulary set for feature interpretations and designing a mutual information-based objective.<n>We propose two runtime steering strategies that adjust the learned feature activations based on their corresponding explanations.
  arXiv  Detail & Related papers  (2025-02-21T16:36:42Z)
• What You See Is Not Always What You Get: An Empirical Study of Code Comprehension by Large Language Models [0.5735035463793009]
  We investigate the vulnerability of large language models (LLMs) to imperceptible attacks, where hidden character manipulation in source code misleads LLMs' behaviour while remaining undetectable to human reviewers.<n>These attacks include coding reordering, invisible coding characters, code deletions, and code homoglyphs.<n>Our findings confirm the susceptibility of LLMs to imperceptible coding character attacks, while different LLMs present different negative correlations between perturbation magnitude and performance.
  arXiv  Detail & Related papers  (2024-12-11T04:52:41Z)
• Contrastive Instruction Tuning [61.97704869248903]
  We propose Contrastive Instruction Tuning to maximize the similarity between semantically equivalent instruction-instance pairs. Experiments on the PromptBench benchmark show that CoIN consistently improves LLMs' robustness to unseen instructions with variations across character, word, sentence, and semantic levels by an average of +2.5% in accuracy.
  arXiv  Detail & Related papers  (2024-02-17T00:09:32Z)

This list is automatically generated from the titles and abstracts of the papers in this site.

This site does not guarantee the quality of this site (including all information) and is not responsible for any consequences.