Decomposing Wrong-Consensus Agreement in LLM Self-Consistency: A GPT-4.1 Case Study

• URL: http://arxiv.org/abs/2608.18795v1
• Date: Wed, 19 Aug 2026 10:50:15 GMT
• Title: Decomposing Wrong-Consensus Agreement in LLM Self-Consistency: A GPT-4.1 Case Study
• Abstract summary: Majority voting over multiple LLM samples is widely used to raise answer accuracy, yet its gain varies erratically.<n>This paper gives a quantitative account of this failure.
• Score: 1.2716872085463884
• License: http://creativecommons.org/licenses/by/4.0/
• Abstract: Majority voting over multiple LLM samples is widely used to raise answer accuracy, yet its gain varies erratically: on hard questions it can even backfire. This paper gives a quantitative account of this failure. A pluralistic agreement index Gamma is defined as the expected fraction of the samples of a wrong run that agree with the consensus, normalized by a reference scale d=(1-p)/(C-1), and is decomposed into a mechanical component (what a vote delivers given only a per-case answer preference) and a preference-unexplained residual. The mechanical null is difficulty-matched and leak-free: each case is resimulated at its own accuracy and option preference, estimated from the case's other runs, so no run predicts its own agreement. On GPT-4.1 the decomposition shows benchmark-associated direction (an observational ordering over n=4 cells per benchmark, not a significance claim). On multiple-choice GPQA-Diamond, the per-case answer preference explains 81-93% of the held-out test-run agreement index: the shared-bias-dominates account over-claims here, because a wrong but attractive option the whole cohort latches onto is captured by the per-case preference channel (whether that preference is induced by shared training bias is not identified). On open-domain AIME, the mechanical preference explains only 59-78% (21-29% if shrunk to pure noise), and a preference-unexplained residual of 1.56-2.80 Gamma units survives, which a run-level preference-heterogeneity reference more than absorbs (1.4-2.1). A self-consistency backfire on hard questions is reproduced (binned voting gap down to -0.09, coupled CI [-0.12,-0.07]), and the highest-agreement bin reaches an accuracy of only 0.42-0.83, a 1.2-3.6x lift over base rate: agreement is graded evidence, not certification. No new voting method is proposed; code and evidence are committed and reproducible.

Related papers

• Symmetry-Aware Likelihood-Orbit Aggregation for Selective Left-Right Claim Verification [14.108900142001628]
  We introduce Relation-Orbit, a closed-form contrast that assigns eight normalized likelihoods to query-supporting and counterfactual roles.<n>On VSR and GQA across four frozen vision-language models, Relation-Orbit yields higher mean test coverage at a 10% selective-risk calibration target.
  arXiv  Detail & Related papers  (2026-09-15T11:19:03Z)
• Optimal Stratified Allocation for Rare-Event Onset Forecasting in Dependent Sequences [2.2632368327435732]
  We study estimation of total risk from a subsample K n under designs allocating K0 and K1 draws to the two strata.<n>We derive the exact finite-population variance of the weighted risk estimator under class-conditional sampling without replacement.
  arXiv  Detail & Related papers  (2026-09-03T19:38:01Z)
• When Many Answers Are Valid, Voting Fails: Symbolic Verification for Best-of-K Causal Reasoning in LLMs [42.32694162421078]
  We introduce CALVER, a training-free symbolic verifier that scores structured traces against Pearl's causal criteria.<n>On CLEAR find-one-valid queries that admit multiple graph-valid answers, CALVER reaches 42.1% where plurality, a reward model, an LLM judge, and model confidence remain near 30%.
  arXiv  Detail & Related papers  (2026-08-04T11:45:46Z)
• UpliftBench: Revealing Outcome-Regime and Objective Mismatch in Uplift Evaluation [0.0]
  UpliftBench evaluates 12 uplift estimators under an outer-test-isolated, multi-objective protocol.<n>We show the disagreement is substantially about metrics, not models.
  arXiv  Detail & Related papers  (2026-08-02T00:51:44Z)
• Knowledge before Reasoning: EC-Reason-Bench, a Training-Free Diagnostic Benchmark for LLM Enzyme Classification [36.44850004100564]
  We propose EC-Reason-Bench, a training-free, diagnostic evaluation protocol built to answer two questions.<n>First, external knowledge is decisive and must precede reasoning.<n>Second, in closed-book settings, whether cascading and chain-of-thought help or hurt depends on a model's tendency to abstain.
  arXiv  Detail & Related papers  (2026-07-29T02:16:09Z)
• K-ABENA: K-Adaptive Backpropagation with Error-based N-exclusion Algorithm : (Compensated Loss-Based Sample Exclusion with Unbiased Gradient Estimation) [0.0]
  K-ABENAK-Adaptive Backpropagation with Error-based N-exclusion is a selective computation framework.<n>It reduces per-it computation training cost by excluding a fraction of low-loss ("minor") observations.<n>It collapses to 0.386 accuracy under label limitation (baseline: 0.832) and to 0.53 AUC under extreme imbalance.
  arXiv  Detail & Related papers  (2026-07-07T06:58:51Z)
• Jury Duty: Calibration and Orientation Failures in MLLM-as-a-Judge Under Cultural Ambiguity [43.23157850664282]
  We introduce VOIR DIRE, a multimodal benchmark of 626 culturally paired image-prompt artifacts spanning U.S. and mainland Chinese contexts.<n>Across six MLLMs, the bias decomposes into two failures: a positivity-floor calibration failure and an orientation failure.<n>We recommend reporting alignment against each reference pool separately and treating cross-pool divergence as a judge property.
  arXiv  Detail & Related papers  (2026-06-12T15:53:40Z)
• Let the Results Speak: A Replication-First Paradigm for LLM Behavioral Benchmarking [22.825786049667602]
  We propose a replication-first paradigm to anchor validity to a single human-rater consensus.<n>We certify the instrument via four properties -- reliability across K runs, cross-instrument replication across architecturally distinct judges, historical-footprint calibration via judges from earlier training cohorts, and pre-registered prediction.<n>We test it on emotional accompaniment by letting the self-evolve data-driven across: the dimensions are not pre-stipulated and the procedure stabilizes to a 9-dimension set.
  arXiv  Detail & Related papers  (2026-05-27T03:41:11Z)
• Distributional Energy-Based Models for Uncertainty-Aware Structured LLM Reasoning [40.342912574072024]
  Large Language Models produce structured outputs such as travel plans or code solutions.<n>Individual reasoning steps may appear correct while the output as a whole violates budgets, fails test cases, or contradicts earlier deductions.<n>We propose a learned quality scorer with deterministic analytical constraint penalties for verifying structured LLM outputs.
  arXiv  Detail & Related papers  (2026-05-15T17:08:27Z)
• Pause and Reflect: Conformal Aggregation for Chain-of-Thought Reasoning [8.024041325202612]
  Chain-of-thought (CoT) reasoning with self-consistency improves performance by aggregating multiple sampled reasoning paths.<n>We introduce a conformal procedure for CoT reasoning that directly addresses aggregation uncertainty.<n>Our approach replaces majority voting with weighted score aggregation over reasoning paths and calibrates an abstention rule using conformal risk control.
  arXiv  Detail & Related papers  (2026-05-13T20:33:59Z)
• The Verification Tax: Fundamental Limits of AI Auditing in the Rare-Error Regime [0.0]
  Most cited calibration result in deep learning -- post-temperature-scaling ECE of 0.012 on CIFAR-100 -- is below the statistical noise floor.<n>We prove this is not a failure of the experiment but a law: the minimax rate for estimating calibration error with model error rate epsilon is Theta((Lepsilon/m)2/3), and no estimator can beat it.
  arXiv  Detail & Related papers  (2026-04-14T16:48:24Z)
• Learning from Emptiness: De-biasing Listwise Rerankers with Content-Agnostic Probability Calibration [76.08899010904652]
  CapCal is a training-free framework that mechanically decouples positional bias from ranking decisions.<n>It achieves superior performance among training-free methods while preserving single-pass efficiency.
  arXiv  Detail & Related papers  (2026-04-11T10:47:22Z)
• How Independent are Large Language Models? A Statistical Framework for Auditing Behavioral Entanglement and Reweighting Verifier Ensembles [46.63622714488747]
  Shared pretraining data, distillation, and alignment pipelines can induce hidden behavioral dependencies, latent entanglement.<n>In practice, this manifests as correlated reasoning patterns and synchronized failures.<n>We develop a statistical framework for auditing behavioral entanglement among black-box language models.
  arXiv  Detail & Related papers  (2026-04-08T23:32:06Z)
• From Calibration to Refinement: Seeking Certainty via Probabilistic Evidence Propagation for Noisy-Label Person Re-Identification [40.73759251488672]
  Existing noise-robust person Re-ID methods rely on loss-correction or sample-selection strategies using softmax outputs.<n>We propose the CAlibration-to-REfinement (CARE) method, a two-stage framework that seeks certainty through probabilistic evidence propagation from calibration to refinement.<n>In the refinement stage, we design the evidence propagation refinement (EPR) that can more accurately distinguish between clean and noisy samples.
  arXiv  Detail & Related papers  (2026-02-26T15:50:15Z)
• From Noisy Traces to Stable Gradients: Bias-Variance Optimized Preference Optimization for Aligning Large Reasoning Models [90.45197506653341]
  Large reasoning models generate intermediate reasoning traces before producing final answers.<n> aligning LRMs with human preferences, a crucial prerequisite for model deployment, remains underexplored.<n>A common workaround optimized a single sampled trajectory, which introduces substantial gradient variance from trace sampling.
  arXiv  Detail & Related papers  (2025-10-06T17:58:01Z)

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.