Search arXiv⌕ Search
Home/Archive/arXiv

arXiv · 2606.05183

The Granularity Gap: A Multi-Dimensional Cross-Generational Audit of Sycophancy in Gemini Models

Patrick Keough
View original source ↗Read source document ↗Export metadata ↓

Abstract

Pass/fail safety evaluation reports whether a model refused. It does not report how far a model went to please the user, and we show these are close to different measurements. We audited sycophancy across three Gemini generations, scoring N=8,830 responses from 8 model variants on 350 adversarial prompts in 7 categories under 3 guardrail conditions, on continuous 1-5 scales for sycophancy, truthfulness and refusal. The judge's own refuse-or-comply verdict explains 29% of the variance in its own sycophancy scores. We term the remainder the Granularity Gap, and it does not close under recalibration: the cut point already in use is the best available on the refusal axis, and no function of that axis explains more than 35%. Reading what four judges wrote while scoring shows why. On a quarter to a third of votes they record that the prompt asked for nothing harmful, almost never in the two categories that solicit a harmful act and up to half the time in the five that do not. A verdict built on refusal has nothing to grade there. Three findings follow. Sycophancy co-occurs with degraded judged truthfulness (rho=0.40), a coupling that strengthens across generations. Capability moved and resistance did not: Gemini 2.0 Flash scores 1.43 and Gemini 3.0 Pro Preview 1.42, with a sharp Gen 2.5 regression between them. And a single direct instruction outperforms an elaborate reasoning protocol in seven of eight variants, cutting mean severity in the most vulnerable category by 60.9%. We evaluate one judge's verdict, not a deployed safety classifier. We release the prompt set, the rubric, and 10,792 per-vote judge scores with their written reasoning.

Explore related subjects

cs.CLcs.AIcs.HC

Keep this discovery

BibTeXRIS

Patrick Keough. 2026-08-28. The Granularity Gap: A Multi-Dimensional Cross-Generational Audit of Sycophancy in Gemini Models. https://arxiv.org/abs/2606.05183

Cite the original work for its findings. Save a collection to share your selection of sources.

Discover connections

cs.AI subjectcs.CL subjectcs.HC subject

Connections use source metadata and explicit phrase matches, not verified experimental comparisons.

KEEP EXPLORING

Related discoveries

arXiv2026

AI Models Can Predict and Collaboratively Modulate Human Memory Search

Large language models (LLMs) exhibit unprecedented natural language generation and many text-based problem-solving capabilities. Indeed, in many language-based tasks, for example routine coding, these artificial intelligence models have reduced, or even eliminated, the need for human input. But rather than replacing human cognitive effort, LLMs may instead serve as cognitive tools to extend human abilities, particularly when they are engaged in a task requiring open-ended conceptual exploration and creative ideation. However, we are yet to understand how these models may enhance such generative human cognitive abilities in human--AI interactions. In this study, we explore and evaluate the ability of LLMs to follow and enhance human mental trajectories during semantic memory search. To test this, we use the semantic fluency task (SFT), a classic cognitive paradigm requiring generative semantic memory retrieval that has long served to characterize convergent and divergent thinking in humans. We demonstrate that an LLM's abilities to track and predict human memory trajectories in this task exceed those of other humans.

cs.CL↗
arXiv2026

Why It Hurts: Identifying the Drivers of Negative Thoughts in Emotional Support Conversations

Large Language Models (LLMs) are increasingly used for emotional support tasks, such as negative thought reframing. This task relies on modifying cognitive appraisals, the subjective interpretation of events that elicit negative emotions, which is typically conceptualized along multiple discrete dimensions. Current LLM-based frameworks model cognitive appraisal by exhaustively evaluating all possible dimensions, but they fail to account for the varying saliency of these dimensions across different contexts. In this work, we investigate a vital yet overlooked question: "Can LLMs infer the salient appraisal dimensions from emotional support conversations?" To address this question, we introduce the AppraiSal benchmark, containing 996 emotional support conversations with human-annotated mental states, including salient cognitive appraisal dimensions. Furthermore, we propose PRISM, a multi-agent probabilistic framework grounded in Bayesian Inverse Planning, designed to improve LLMs' ability to identify context-specific appraisal dimensions. Experimental results show that PRISM brings improvements to LLMs across various sizes, particularly in identifying the most salient appraisal dimensions.

cs.HC↗
arXiv2026

Toward Cultural Alignment: Human-Centered Evaluation of Multimodal AI Stories Across Five African Communities

In this paper, we examine how well AI-generated multimodal stories align with the lived practices, relationships, language, values, and visual expectations of the communities they represent. We conduct a community-grounded mixed-methods evaluation with 19 culture representatives across five African communities, combining quantitative annotations with qualitative focus group discussions. We find that cultural alignment depends not simply on recognizable cultural markers, but on how those markers fit social, linguistic, procedural, and visual context. From these evaluations, we develop a taxonomy of cultural alignment comprising five broader cultural marker categories and eight recurring mechanisms of misalignment. We additionally evaluate five multimodal LLM judges to examine whether automated evaluation can approximate community-grounded judgments at scale. Judge reliability and score calibration vary substantially across communities, with no single judge performing consistently across all five settings. These findings motivate community-calibrated evaluation pipelines in which automated judges are validated against community judgments to determine where they can be trusted and where human review remains necessary.

cs.CL↗