Tag Questions and the Generational Reversal of Sycophancy Across 45 Language Models

• URL: http://arxiv.org/abs/2607.23976v1
• Date: Mon, 27 Jul 2026 04:02:00 GMT
• Title: Tag Questions and the Generational Reversal of Sycophancy Across 45 Language Models
• Abstract summary: Appending a two-word confirmation tag to a decision question changes whether a language model endorses the choice.<n>We measure this tag effect on 20 frozen, ground-truth-free decisions.<n>Across 45 models the effect spans +32% to -32% -- a 64-point swing on one word.
• Score: 3.7734715043126488
• License: http://creativecommons.org/licenses/by/4.0/
• Abstract: Appending a two-word confirmation tag to a decision question -- "Is X the better choice?" versus "X is the better choice, right?" -- changes whether a language model endorses the choice. We measure this tag effect on 20 frozen, ground-truth-free decisions between two defensible options, counterbalanced so a model's own preferences cancel, scored by exact match on clamped yes/no replies -- no LLM judge, no embeddings. Across 45 models the effect spans +32% to -32% -- a 64-point swing on one word -- with 5 models significantly sycophantic and 17 significantly resistant (BH-FDR q=.10). The sign is a clock: within model families the effect crosses from positive to negative as generations advance (GPT +4 to -28; Claude +7 to -32; Qwen and Grok likewise), roughly -6 points per year, a reversal robust to vendor tier; one lineage (DeepSeek) never crosses, and two releases during the study window (Claude Opus 5, Gemini 3.6 Flash) land on the trend out-of-sample. A full-panel ablation localizes the resistance as a double dissociation: a synonym tag reproduces each model's response almost exactly (r=0.89), while planting the same preference without a tag produces resistance in no resistant model (stance effects +6 to +49; r=0.23 with tag effects). The resistance is keyed to the surface construction of a tacked-on agreement bid, not the user's stance -- a pattern-match, not a principle. And the tag's polarity matters more than its presence: swap one word -- "X is the better choice, maybe?" -- and agreement rises above the neutral baseline in 45 of 45 models (+19.6 points), with ten models affirming both mutually exclusive options at 90-100%. Agreement tracks how sure the user sounds, in opposite directions at the two poles. The instrument is one word, one dollar, and judge-free; run per release, it reads the field's anti-sycophancy training directly off model behavior.

Related papers

• Should I Be Polite to My LLM Relevance Judge? Tone as a Severity Operating-Point Shift [7.837842601637295]
  Tone affects calibration-based agreement more than ranking outcomes.<n>Where tone changes agreement, the results are more consistent with a shift in the judge's severity operating point.
  arXiv  Detail & Related papers  (2026-09-09T04:40:14Z)
• Marginal Fidelity Does Not Establish User Simulation in Demographic Synthetic Survey Panels: Response Contracts, Support Collapse and Conditioning Failure [51.736723807086385]
  Population-marginal agreement is evidence about an elicitation contract and an estimand obtainable without simulated respondents, not evidence of individual simulation.<n>On nine aligned model-battery pairs, a no-persona population-prevalence query averages 6.27 MAE versus 12.39 for committed panels and wins all nine comparisons.
  arXiv  Detail & Related papers  (2026-09-07T10:22:22Z)
• GradeTrap: Authority Cues in Images Shift VLM Judgments Despite Explicit Instructions to Ignore Them [0.0]
  We introduce GradeTrap, a controlled evaluation that places two social cues in direct conflict.<n>A student answer should attract sycophantic agreement, and a conflicting answer attributed to a peer, teacher, or official answer key should attract authority-based deference.<n>We test the models on 60 synthetic real-world trade-off scenarios.
  arXiv  Detail & Related papers  (2026-09-05T12:30:39Z)
• Right Frame, Wrong Rule: Cultural Cues Expose the Financial Knowledge Gap They Were Meant to Close [57.963463622270275]
  We study normative pluralism in Islamic finance using a four-choice taxonomy.<n>A cultural cue steers a model toward one framework, but the model selects an incorrect answer within that framework.
  arXiv  Detail & Related papers  (2026-09-01T09:46:53Z)
• TrustRoboReward: Preference-Ordered Isotonic Score Editing for Multi-Paradigm Robot Reward Models [40.42873860258522]
  We propose TrustRoboReward, a multi-paradigm reward modeling framework equipped with Preference-Ordered Isotonic Score Editing (POISE)<n>POISE rectifies pointwise scores and eliminates cross-paradigm reversal conflicts unresolved by TrustJudge.<n>Qwen3-VL-4B trained with POISE achieves an overall reward score of 77.96%, nearly matching GPT-5-mini (78.09%, gap 0.13%) and outperforming the strongest RoboReward-4B baseline by 10.13%.
  arXiv  Detail & Related papers  (2026-08-09T05:25:22Z)
• When AI Writes, Who Gets Cited? Evidence of Citation Monoculture Across Language Models [45.91367912142692]
  We show that different models may still select the same narrow subset, producing citation monoculture without any single citation being wrong.<n>Eleven models from three vendors choose at most ten papers from uniformly random panels of thirty, with real titles and abstracts but fabricated authors, reassigned years, and hidden venues and citation counts.
  arXiv  Detail & Related papers  (2026-08-03T14:27:46Z)
• Conformity Mitigations in Large Language Models Lie on a Single Resistance-Receptivity Frontier [0.05055376568994175]
  A wrong majority reverses 22.8% of a model's correct MMLU answers, 54.8% on GPQA and 71.0% on SimpleQA.<n>We pair it with Receptivity, the rate at which a model adopts a correct peer answer after initially answering incorrectly.<n> Reflection, the strongest published method, gains 7.9 points of MMLU Resistance and gives up 15.3 of Receptivity.
  arXiv  Detail & Related papers  (2026-08-03T11:12:38Z)
• OptimismBench: Forecasting Bias and the Alignment Effect in Language Model Judgment [0.9821874476902969]
  We introduce OptimismBench, which detects directional bias with inverted pairs.<n>We show post-training sets the sign of the bias, with opposite shifts in different families.<n>We release 3,870 items across 10 languages for per-model directional-bias auditing.
  arXiv  Detail & Related papers  (2026-07-29T14:38:55Z)
• Logical Judgments Under Pressure: Diagnosing Syllogistic Stability with Learned Soft Prefixes [1.1844977816228046]
  We characterize how learned contextual pressure can override correct judgments and expose limits in a model's logical stability.<n>Across Qwen3.6-35B-A3B MoE, Qwen3-8B, and Gemma, learned prefixes redirect many correct answers and remain effective across unseen forms and interface changes.
  arXiv  Detail & Related papers  (2026-07-20T17:58:05Z)
• Abliteration Is Not a Scalpel: Off-Target Effects of Refusal Removal on Decision Disposition Across Model Families [51.56484100374058]
  Abliteration - deleting a model's refusal direction from its weights - is the standard recipe behind popular "uncensored" open-weight models.<n>As a disposition probe we use 21,600 decisions under uncertainty, replayed through a frozen pipeline so the decision-layer model is the only variable.<n>Three effects replicate across both families (weeks-clustered CIs excluding zero)<n>A fourth effect reverses sign: the same operation makes Gemma-abliterated less confident and Qwen-abliterated more (family CIs non-overlapping)
  arXiv  Detail & Related papers  (2026-07-19T22:27:00Z)
• The One-Word Census: Answer-Choice Conformity Across 44 Language Models [3.7734715043126488]
  We characterize convergence of 31 single-turn prompts, each naming a category with many valid one-word answers.<n>We score each model by answer-choice surprisal: the average $-log2$ probability of its answers under the pooled answers of all other models.<n> Persona- and community-tuned models are the most divergent; the newest mainline flagships are the most conformist.
  arXiv  Detail & Related papers  (2026-07-14T14:12:05Z)
• The yes-no bias of large language models reflects answer order and wording, not shifts in moral judgment [3.7193230342956056]
  A growing literature reports such judgments shifting under logically irrelevant changes of wording.<n>In a yes/no question the word "no" is at once logical verdict, lexical token, and last-printed option.<n>We introduce a psychometric battery that separates these: crossed symmetrization across a corpus of question forms.
  arXiv  Detail & Related papers  (2026-07-06T18:37:43Z)
• CAVEWOMAN: How Large Language Models Behave Under Linguistic Input and Output Compression [77.41779716950387]
  We present Cavewoman, a two-channel evaluation protocol that scores every generation on task accuracy, realized per-item cost, and reference-text agreement.<n>We evaluate eight models on five datasets at five reduction levels, with both channels measured on the same items.
  arXiv  Detail & Related papers  (2026-06-23T02:56:48Z)
• DART: Draft-Agreement Routing for Training-Free Adaptive Thinking Budgets in Hybrid Reasoning Models [45.71412371299232]
  We introduce DART, a training-free routing framework that samples two cheap no-think drafts.<n>DART accepts direct answering when the drafts agree, and predicts a thinking budget from draft entropy when they disagree.<n>On math reasoning, accuracy improves by up to $+$9.0 points on Olympiad-level problems while thinking tokens drop 15-69%.<n>On code reasoning under execution-based equivalence, accuracy improves by up to +22.5 points while thinking tokens drop 51-63%.
  arXiv  Detail & Related papers  (2026-06-22T11:24:59Z)
• CoEval: Ranking Language Models for Custom Tasks Without Labeled Data or Trustworthy Benchmarks [47.027290803102666]
  We present CoEval, an open framework that supplies a trustworthy, task-specific signal through ensemble self-evaluation.<n>A pool of models rotates through all three roles, teacher, student, and judge, to generate a fresh, contamination-free benchmark.
  arXiv  Detail & Related papers  (2026-06-02T13:41:43Z)
• The Chain Holds, the Answer Folds: Trace-Answer Dissociation in Reasoning Models Under Adversarial Pressure [8.86745721473138]
  Reasoning models are evaluated on single-turn benchmarks but deployed in multi-turn dialogue, where users push back on correct answers.<n>We call this unfaithful capitulation (UC) and isolate it with a $2times 2$ latent-versus-behavioral framework that flip-rate metrics and single-turn probes both miss.
  arXiv  Detail & Related papers  (2026-05-27T20:41:08Z)
• Why Models Know But Don't Say: Chain-of-Thought Faithfulness Divergence Between Thinking Tokens and Answers in Open-Weight Reasoning Models [0.0]
  Extended-thinking models expose a second text-generation channel ("thinking tokens") alongside the user-visible answer.<n>This study examines 12 open-weight reasoning models on MMLU and GPQA questions paired with misleading hints.
  arXiv  Detail & Related papers  (2026-03-27T13:39:05Z)
• Self-Harmony: Learning to Harmonize Self-Supervision and Self-Play in Test-Time Reinforcement Learning [48.14470449860784]
  Test-time reinforcement learning (TTRL) offers a label-free paradigm for adapting models using only synthetic signals at inference.<n>We introduce Self-Harmony, a framework built on a simple intuition.
  arXiv  Detail & Related papers  (2025-11-03T03:34:34Z)
• Causal Understanding by LLMs: The Role of Uncertainty [43.87879175532034]
  Recent papers show LLMs achieve near-random accuracy in causal relation classification.<n>We investigate whether pretraining exposure to causal examples improves causal understanding.
  arXiv  Detail & Related papers  (2025-09-24T13:06:35Z)
• RLCD: Reinforcement Learning from Contrastive Distillation for Language Model Alignment [121.45689748315125]
  Reinforcement Learning from Contrastive Distillation (RLCD) is a method for aligning language models without using human feedback. RLCD creates preference pairs from two contrasting model outputs, one using a positive prompt designed to encourage following the given principles, and one using a negative prompt designed to encourage violating them. We then use the preference pairs to train a preference model, which is in turn used to improve a base unaligned language model via reinforcement learning.
  arXiv  Detail & Related papers  (2023-07-24T17:23:22Z)

This list is automatically generated from the titles and abstracts of the papers in this site.

This site does not guarantee the quality of this site (including all information) and is not responsible for any consequences.