mlsanityData to 14 Oct 2026

Suppressing Pressure, Amplifying Evidence: Self-Guided Attention Steering to Mitigate Sycophancy and Stubbornness

arXiv 2610.04329 ↗First posted 3 Oct 2026cs.AI cs.CLcited by 0

Not on the map. Why?

Abstract (arXiv)

Reliable language models should resist unsupported user pressure while effectively using objective contextual information. However, models may exhibit sycophancy by yielding to unsupported user pressure or contextual stubbornness by failing to update their answers when relevant contextual information warrants revision. Evaluating interventions for these failures separately can obscure whether mitigating one failure exacerbates the other. To assess this trade-off, we introduce CoPE-Bench with six conditions per question: a neutral baseline, correct or incorrect user pressure, contextual information consistent with or conflicting with the neutral answer, and a joint condition combining incorrect claims with conflicting contextual information. To regulate the influence of user pressure and contextual information, we propose SPAE (Suppressing Pressure, Amplifying Evidence), a training-free framework that uses the model's own judgments to identify relevant tokens, suppressing user pressure and amplifying contextual information through token-level attention steering. On average across five backbones, SPAE reduces pressure following by 18.8 percentage points and increases joint-condition updating by 5.5 percentage points relative to the strongest baseline in the main comparison. In two-turn dialogue, it improves joint-condition updating by an average of 13.2 percentage points over the strongest prompting baseline. The source data and codes can be found at https://github.com/03Grant/sycophancy-and-stubbornness.

Key facts. No citing papers in the map yet (data through 14 Oct 2026). Builds on 28 of its 33 references that are in the map. First posted on arXiv 3 Oct 2026.

Builds on

28 of 33 references.

References · 28

  1. 2021TruthfulQA: Measuring How Models Mimic Human Falsehoodscited by 2,975
  2. 2023Steering Llama 2 via Contrastive Activation Additioncited by 1,085
  3. 2022Discovering Language Model Behaviors with Model-Written Evaluationscited by 836
  4. 2021Entity-Based Knowledge Conflicts in Question Answeringcited by 356
  5. 2023Adaptive Chameleon or Stubborn Sloth: Revealing the Behavior of Large Language Models in Knowledge Conflictscited by 341
  6. 2023Trusting Your Evidence: Hallucinate Less with Context-aware Decodingcited by 327
  7. 2026Gemma 4 Technical Reportcited by 259
  8. 2023Inverse Scaling: When Bigger Isn't Bettercited by 207
  9. 2025SycEval: Evaluating LLM Sycophancycited by 108
  10. 2025Measuring Sycophancy of Language Models in Multi-turn Dialoguescited by 104
  11. 2023Tell Your Model Where to Attend: Post-hoc Attention Steering for LLMscited by 91
  12. 2023System 2 Attention (is something you might need too)cited by 89
16 more
  1. 2023Context-faithful Prompting for Large Language Modelscited by 80
  2. 2023Are You Sure? Challenging LLMs Leads to Performance Drops in The FlipFlop Experimentcited by 77
  3. 2024From Yes-Men to Truth-Tellers: Addressing Sycophancy in Large Language Models with Pinpoint Tuningcited by 54
  4. 2024AdaCAD: Adaptively Decoding to Balance Conflicts between Contextual and Parametric Knowledgecited by 37
  5. 2025Taming Knowledge Conflicts in Language Modelscited by 30
  6. 2026Sycophancy Hides Linearly in the Attention Headscited by 23
  7. 2025Spotlight Your Instructions: Instruction-following with Dynamic Attention Steeringcited by 21
  8. 2024Model Tells Itself Where to Attend: Faithfulness Meets Automatic Attention Steeringcited by 17
  9. 2025MONICA: Real-Time Monitoring and Calibration of Chain-of-Thought Sycophancy in Large Reasoning Modelscited by 12
  10. 2025Sycophancy Mitigation Through Reinforcement Learning with Uncertainty-Aware Adaptive Reasoning Trajectoriescited by 11
  11. 2026Towards Reliable LLM Evaluation: Correcting the Winner's Curse in Adaptive Benchmarkingcited by 7
  12. 2025Pointing to a Llama and Call it a Camel: On the Sycophancy of Multimodal Large Language Modelscited by 5
  13. 2025Assessing and Mitigating Medical Knowledge Drift and Conflicts in Large Language Modelscited by 4
  14. 2026MIRA: A Bilingual Benchmark for Medical Information Response Auditcited by 2
  15. 2026Sycophancy Suppression Can Impair Rational Updating: Anti-Sycophancy Should Preserve the Ability to Updatecited by 2
  16. 2026Equitable System-Prompt Selection via Constrained Mixed-Strategy GroupDROcited by 1

Built on by

Not cited yet.