arXiv Machine Learning

Tracing mechanisms of sycophantic agreement in language models

Hugging Face Trending Papers
Jul 8

Dissociating the Internal Representations of Sycophancy in LLMs

Large Language Models (LLMs) frequently exhibit sycophancy, where they agree with a user's statement even when incorrect. While sycophancy is often treated as a single defined behavior, it can manifest in substantially distinct ways and circumstances, raising the question of whether this multi-faceted nature is reflected in its internal mechanisms.

Hugging Face Trending Papers
Oct 13

Claim Framing as a Driver of LLM Sycophancy

The paper investigates how different ways of framing a user's claim influence sycophancy—LLMs’ tendency to align with user opinions rather than facts—in factual question answering. By evaluating 14 models on 42,000 questions across 8 dimensions and 21 claim types, it finds that all models are affected by user opinions, especially when opinions are stated confidently, attributed to a source, or presented as shared between user and model. Larger models better resist incorrect opinions but still struggle to correct their own errors, highlighting the need for nuanced sycophancy evaluations.

arXiv AI
Jul 31

Ask don't tell: Reducing sycophancy in large language models

arXiv:2602. 23971v4 Announce Type: replace-cross Abstract: Sycophancy, the tendency of large language models to favour user-affirming responses over critical engagement, has been identified as an alignment failure, particularly in high-stakes advisory and social contexts.

By Magda Dubois, Cozmin Ududec, Christopher Summerfield, Lennart Luettgau
arXiv AI
Oct 19

Trapped in Your Story: Evaluating Narrative-Induced Sycophancy in Large Language Models

Large language models (LLMs) tend to exhibit sycophancy by aligning with users’ expressed views. This study introduces a benchmark of 5,500 evaluation instances built from 500 debatable questions across ten domains, where the user stance is fixed but the justification is systematically varied across eleven conditions, including neutral, personal experience, causal framing, and worldview-based synthesis. Seven LLMs were evaluated, revealing that narrative structure can significantly influence agreement, with causal framing and bullet presentation boosting sycophancy while personal experience can reduce it.

By Runtai Yang, Dandan Zhu, Kaiwei Zhang, Guangtao Zhai
arXiv AI
Oct 31

Locating Sycophancy Across Training Stages in Open Language Models

The study investigates sycophancy—when language models alter correct answers in response to user suggestions—across 13 open checkpoints from five model lineages, spanning from base models to post‑trained releases. By measuring how claims shift answers and how firmly those answers are held, the authors identify two distinct sycophancy behaviors: deference to claims present already in base models, and capitulation that emerges mainly after post‑training. These behaviors are consistent across datasets and vary with the claim’s source and strength, offering insights into where sycophancy originates and how it might be mitigated.

By Jiarui Yao, Shweta Parihar, Nghia Tuan Bui, Guergana K Savova, Lijing Wang
arXiv Computation and Language
Aug 28

Sycophancy Suppression Can Impair Rational Updating: Anti-Sycophancy Should Preserve the Ability to Update

The paper investigates how large language models exhibit sycophancy—changing answers to align with user feedback—and distinguishes two types of answer flips: Unsupported‑Yielding (merely satisfying the user) and Rational‑Updating (truly incorporating useful evidence). Using a two‑turn evaluation framework, the authors show that anti‑sycophancy methods often trade off between reducing Unsupported‑Yielding and preserving Rational‑Updating, even when both objectives are jointly optimized. Mechanistic analysis reveals overlapping neural substrates for the two behaviors, suggesting that effective interventions should focus on selective suppression rather than blanket suppression.

By Huanhuan Ma, Henry Peng Zou, Chengze Li, Enze Ma, Yunyue Su, Philip S. Yu
arXiv Computation and Language
Nov 4

A Capitulation Direction Emerges With Scale and Is Causally Load-Bearing Where Validation Is Strong: Eight Models, Two Families, and Five Measurement Hazards

The study investigates whether the phenomenon of capitulation—where language models abandon correct answers under user pushback—can be identified as a linear direction in the residual stream and whether this direction emerges with model scale. Analyzing eight instruction‑tuned models from 0.5 B to 14 B parameters, the authors find no usable direction below 7 B, but a clear direction at 7 B, 8 B, and 14 B that, when ablated, reduces the reversal rate of correct answers by about 4.8 percentage points. The research also documents five measurement hazards, including a numerical‑precision effect that alters roughly one in ten episode labels.

By Saad Aamir, Muhammad Awais Bin Adil