Which Knowledge Should LLMs Trust? Uncovering a Generalizable Steering Direction Across Knowledge Conflicts

arXiv:2610.20067 · cs.CL, cs.AI · Submitted 2026-10-14 · Read on arXiv

cs.CL, cs.AI

Submitted: 2026-10-14

Updated: 2026-10-14

Code: https://github.com/AheadOFpotato/knowledge_selection_steering

License: http://arxiv.org/licenses/nonexclusive-distrib/1.0/

Terminology

• Steering — 1× in this paper · explained in 2 episodes

Sources

• ContextFocus: Activation Steering for Contextual Faithfulness in Large Language Models
• Refusal in Language Models Is Mediated by a Single Direction
• Parameters vs. Context: Fine-Grained Control of Knowledge Reliance in Language Models
• DRAGged into Conflicts: Detecting and Addressing Conflicting Sources in Search-Augmented LLMs
• Benchmarking Large Language Models in Retrieval-Augmented Generation
• Understanding the Interplay between Parametric and Contextual Knowledge for Large Language Models
• Conflict-Aware Soft Prompting for Retrieval-Augmented Generation
• Toy Models of Superposition
• Retrieval-Augmented Generation for Large Language Models: A Survey
• Dissecting Recall of Factual Associations in Auto-Regressive Language Models
• Why So Gullible? Enhancing the Robustness of Retrieval-Augmented Models against Counterfactual Noise
• Cross-model Transferability among Large Language Models on the Platonic Representations of Concepts
• Mistral 7B
• Cutting Off the Head Ends the Conflict: A Mechanism for Interpreting and Mitigating Knowledge Conflicts in Language Models
• Taming Knowledge Conflicts in Language Models
• Inference-Time Intervention: Eliciting Truthful Answers from a Language Model
• Entity-Based Knowledge Conflicts in Question Answering
• Evaluating Very Long-Term Conversational Memory of LLM Agents
• When Not to Trust Language Models: Investigating Effectiveness of Parametric and Non-Parametric Memories
• The Geometry of Truth: Emergent Linear Structure in Large Language Model Representations of True/False Datasets

Related papers

• Exploring Solution Divergence and Its Effect on Large Language Model Problem Solving
• Ishigaki-IDS-Bench: A Benchmark for Generating Information Delivery Specification from BIM Information Requirements
• Subliminal Steering: Stronger Encoding of Hidden Signals
• MedStruct-S: A Benchmark for Key Discovery, Key-Conditioned QA and Semi-Structured Extraction from OCR Clinical Reports
• The End of Transformers? On Challenging Attention and the Rise of Sub-Quadratic Architectures
• Untangling the Mechanisms of Misleading Context in Medical Question Answering