Entity · paper

Partition, Prompt, Aggregate: Statistical Self-Consistency in Language Models

paperactiveprovisionalpartition-prompt-aggregate-statistical-self-consistency-in-language-models-07cc74e6·1 events·first seen Jul 17, 2026

Aliases: Partition, Prompt, Aggregate: Statistical Self-Consistency in Language Models

More like this (12)

Self-Compacting Language Model Agents LESS: Mutual-Stability Sampling for Diffusion Language Models Sound Probabilistic Safety Bounds for Large Language Models The Shibboleth Effect: Auditing the Cross-Lingual Distributional Skew of Large Language Models Automated reproducibility assessments in the social and behavioral sciences using large language models On the Limits of Prompt-Conditioned Language Models as General-Purpose Learners Apparent Psychological Profiles of Large Language Models are Largely a Measurement Artifact Decomposing Factual Sycophancy in Language Models: How Size and Instruction Tuning Shape Robustness Random Language Model Self-Augmenting Retrieval for Diffusion Language Models Beyond Uniform Tokens: Adaptive Compression for Time Series Language Models Construction-Driven Injection: Linguistically-Grounded Edit-Based Code-Mixing Fingerprints for Large Language Models

Recent events (1)

5arXiv · cs.CL·Jul 17, 2026·source ↗

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.

Evaluation and Benchmarking AI Safety Research Partition, Prompt, Aggregate: Statistical Self-Consistency in Language Models