5arXiv cs.CL (Computation and Language)·
LLMs violate statistical self-consistency: the macro fallacy in population-level estimation
A new arXiv paper introduces a framework for testing whether LLM outputs satisfy the law of total probability — specifically, whether prior-weighted conditional estimates over subpopulations aggregate correctly into population-level marginals. Using binary tree partitions and persona prompting across frontier models, the authors find widespread violations of this self-consistency principle. A key finding, termed the 'macro fallacy,' is that estimates reconstructed from fine-grained subpopulation prompts are often more accurate than direct population-level estimates, suggesting models hold relevant subpopulation knowledge but fail to propagate it upward. The work proposes statistical self-consistency as a reference-free, unsaturated evaluation criterion for LLMs.
Related guides (2)
Related events (8)
Benchmarking study finds LLMs fail at counterintuitive probability problems despite strong standard performance
A new arXiv paper evaluates 8 state-of-the-art LLMs on discrete probability problems using two datasets: standard exercises (average accuracy 0.96) and counterintuitive exercises designed to trigger heuristic reasoning (average accuracy 0.59). The authors document token bias causing 20%+ performance drops when canonical problem formulations are disguised, and up to 34% degradation when misleading suggestions are embedded in prompts. The findings argue that current LLMs are not genuine probabilistic reasoners despite their success on advanced math benchmarks.
LLM detection tools can paradoxically increase LLM usage and reduce output quality, paper argues
A new arXiv preprint models LLM detectors as behavioral interventions and shows that imperfect detectors can produce counterintuitive outcomes: users may increase LLM usage and produce lower-quality outputs in response to detection pressure. The authors develop a stylized game-theoretic model of strategic user behavior and empirically validate a predicted 'rise-then-fall' pattern in detectable word frequencies using arXiv abstract data. The work identifies systematic failure modes for LLM detection policies deployed in academic, professional, or platform contexts.
LLMs automate reproducibility assessments in social and behavioral sciences, outperforming human reanalysts
A preprint from arXiv demonstrates that an LLM pipeline can automate reproducibility assessments of published social and behavioral science studies, recovering original effect sizes in 41% of cases (vs. 34% for human reanalysts) and reaching the same qualitative conclusion in 96% of cases (vs. 74% for humans). The study evaluated 76 published studies with predefined claims. The results suggest LLMs could serve as a scalable tool for systematic auditing of empirical research, addressing the resource-intensive nature of traditional reproducibility efforts.
Mathematical framework models linguistic monoculture risk from widespread LLM writing assistance
A new arXiv preprint develops a formal mathematical framework to analyze how widespread LLM-assisted writing may reduce population-level linguistic diversity, a phenomenon the authors term 'linguistic monoculture.' The paper models authors and LLMs as distributions over linguistic features and analyzes three interaction mechanisms—fixed shared models, recursively updated shared models, and personalized models—characterizing equilibria and convergence rates for each. A key finding is that individually rational authors may over-conform relative to the social optimum because they do not internalize the value of their own distinctiveness, creating a negative externality. The 'price of monoculture' can grow without bound when distinctiveness dominates authenticity in the utility model.
Paper challenges LLM expert-level claims by measuring variance and error magnitude in code-based data analysis tasks
A new arXiv paper argues that standard LLM benchmarks overstate model capabilities by focusing on average performance on training-data-adjacent tasks while ignoring response variance and error magnitude. The authors introduce a novel benchmark requiring frontier LLMs to write code for data analysis tasks, comparing results against human expert submissions. Human experts outperformed the frontier LLM on average across multiple metrics and showed lower performance variability. The findings challenge the prevailing narrative that LLMs perform at human-expert level on knowledge economy tasks.
LLMs fail to consistently simulate demographic perspective-taking in hate speech annotation
A new arXiv paper evaluates whether persona-conditioned LLMs can replicate how different demographic groups perceive hate speech, testing three dimensions: inter-group disagreement, in-group sensitivity, and vicarious prediction. No model consistently captures all three dimensions, and performance is highly model-dependent rather than emerging reliably from identity prompts alone. Vicarious prompting with Llama 3.1 provides the closest approximation to human disagreement patterns across demographic axes. The findings have implications for using LLMs as proxies for diverse human annotators in content moderation tasks.
LLMs reproduce human behavioral biases in route choice without explicit prospect-theoretic parameters
A new arXiv preprint investigates whether LLMs can replicate systematic human decision-making biases in route choice scenarios without explicitly specifying cumulative prospect theory (CPT) parameters. The authors design a behavioral evaluation framework and find that LLMs reproduce non-rational choice patterns consistent with CPT effects under uncertainty. The findings suggest LLMs could serve as a scalable substitute for survey-based parameter calibration in agent-based behavioral simulations, addressing a longstanding bottleneck in large-scale human decision modeling.
LLMs fail to reliably self-report adversarial prefill attacks, study finds
A new arXiv paper evaluates whether LLMs can recognize that their own prior responses were elicited by adversarial prefill attacks, testing ten open-weight models (3B–70B) across four safety benchmarks. Models claim intent on prefilled responses only 27.3% of the time on average, and introspective signal is largely mediated by refusal-related reasoning. Three LoRA fine-tuning methods (SFT, GRPO, DPO) improve the intention-probe gap but counterintuitively raise attack success rates on most models, suggesting partial and fragile mitigation. The findings raise concerns about the reliability of LLM self-reports in safety-critical contexts.