Context Copying Modulation: The Role of Entropy Neurons in Managing Parametric and Contextual Knowledge Conflicts

2025/09/12 by Zineddine Tighidet, Tighidet, Zineddine, Andrea Mogini +9 · 1 voice · 1 citation
Computer Science · Neuroscience · Social Sciences · #Computation and Language (cs.CL) #Embodied and Extended Cognition #FOS: Computer and information sciences #Language and cultural evolution #Topic Modeling #cs.CL

paper · pdf · doi:10.48550/arxiv.2509.10663

openalex publication_date 2025/09/12 · arxiv published 2025/09/12 · arxiv updated 2025/09/17 · openalex created_date 2025/10/10 · openalex updated_date 2026/07/28

Abstract

The behavior of Large Language Models (LLMs) when facing contextual information that conflicts with their internal parametric knowledge is inconsistent, with no generally accepted explanation for the expected outcome distribution. Recent work has identified in autoregressive transformer models a class of neurons -- called entropy neurons -- that produce a significant effect on the model output entropy while having an overall moderate impact on the ranking of the predicted tokens. In this paper, we investigate the preliminary claim that these neurons are involved in inhibiting context copying behavior in transformers by looking at their role in resolving conflicts between contextual and parametric information. We show that entropy neurons are responsible for suppressing context copying across a range of LLMs, and that ablating them leads to a significant change in the generation process. These results enhance our understanding of the internal dynamics of LLMs when handling conflicting information.

Citations

• Probing Language Models on Their Knowledge Source
• IRCAN: Mitigating Knowledge Conflicts in LLM Generation via Identifying and Reweighting Context-Aware Neurons
• Confidence Regulation Neurons in Language Models
• Knowledge Conflicts for LLMs: A Survey
• Cutting Off the Head Ends the Conflict: A Mechanism for Interpreting and Mitigating Knowledge Conflicts in Language Models
• Universal Neurons in GPT2 Language Models
• A Glitch in the Matrix? Locating and Detecting Language Model Grounding with Fakepedia
• Adaptive Chameleon or Stubborn Sloth: Revealing the Behavior of Large Language Models in Knowledge Conflicts
• VISIT: Visualizing and Interpreting the Semantic Information Flow of Transformers
• Dissecting Recall of Factual Associations in Auto-Regressive Language Models
• Fine-tuning a Subtle Parsing Distinction Using a Probabilistic Decision Tree: the Case of Postnominal "that" in Noun Complement Clauses vs. Relative Clauses
• A Review on Language Models as Knowledge Bases
• Locating and Editing Factual Associations in GPT
• Transformer Feed-Forward Layers Are Key-Value Memories
• Language Models as Knowledge Bases: On Entity Representations, Storage Capacity, and Paraphrased Queries
• PyTorch: An Imperative Style, High-Performance Deep Learning Library
• Attention Is All You Need

Cited by

• Classification EM-PCA for clustering and embedding
• United Minds or Isolated Agents? Exploring Coordination of LLMs under Cognitive Load Theory

Discussions

• In collaboration with BNP Paribas. Arxiv: arxiv.org/abs/2509.10663 Code: github.com/Zineddine-Ti... Project page: zineddine-tighidet.github.io/projects/pro... [bsky, 0 points, 1 comments]

Related

• Confidence Regulation Neurons in Language Models
• Revisiting Transformers through the Lens of Low Entropy and Dynamic Sparsity
• Mitigating Copy Bias in In-Context Learning through Neuron Pruning
• Representational Curvature Modulates Behavioral Uncertainty in Large Language Models
• Entropy-Lens: The Information Signature of Transformer Computations