English static mirror for SEO/GEO · AI-assisted translation · Read Chinese original

Stochastic Sampling Is Epistemically Shallow: Why Asking an LLM 100 Times Doesn't Reveal Truth

Forum topic · 小凯 · 2026-07-25

Summary

A forum post discusses a paper by Izhar Ali (Rowan University), arXiv:2607.20464, presented at the EIML@ICML 2026 workshop, arguing that self-consistency sampling from a single LLM captures noise, not knowledge. Using Marchenko-Pastur random matrix theory, the author compares a 100-run sampling matrix from Qwen2.5-7B (temperature=1) across 500 questions against a 24-model matrix (temperature=0). The single model's samples show at most one eigenvalue above the noise floor (Tracy-Widom z-score ≤ +0.05), while the diverse model ensemble shows four statistically validated signals — a 'dimensionality gap' of ≤1 versus ≥4. The paper explains this via temperature as a uniform scaling of logits that cannot produce domain-structured variation, and observes models are 'locally committed, globally incoherent' (87.4% of resamples repeat the same wrong answer without cross-question error correlations). Practical implications: self-consistency is fine for per-question selective prediction and hallucination flags, but revealing systematic capability weaknesses requires true model diversity, making ensembles an epistemically different information source rather than a costlier version of repeated sampling.

This post from zhichai.net reviews a paper arguing that repeated sampling from the same LLM cannot substitute for genuine model diversity.

> Paper: Stochastic Sampling is Epistemically Shallow: The Dimensionality Gap Between Temperature Variation and Model Diversity in LLMs > Author: Izhar Ali (Rowan University) > arXiv: 2607.20464 > Venue: EIML@ICML 2026 (2nd Workshop on Epistemic Intelligence in Machine Learning)

Key points

  • Self-consistency under scrutiny. Self-consistency (Wang et al., 2023) samples multiple answers from one LLM and majority-votes them; it works well on benchmarks like GSM8K and underpins confidence estimation, semantic-entropy hallucination detection, etc. The paper asks whether this variation carries structured knowledge signals or mere noise.
  • Two kinds of uncertainty. Per-question uncertainty (does the model disagree with itself on this question?) differs from cross-question structure (are errors systematically correlated across related problems, revealing capability dimensions?). Self-consistency addresses the former but not the latter.
  • Method: Marchenko-Pastur random matrix theory. Correctness matrices (runs × questions) are tested against the MP eigenvalue distribution for pure noise; eigenvalues above the MP edge indicate non-random structure.
  • Experiment. Group A: one LLM (Qwen2.5-7B), 100 samples per question at temperature=1, over 500 questions. Group B: 24 different LLMs, one deterministic run each.
  • Result: the dimensionality gap. Group A yields at most one eigenvalue above the noise floor, and it is weak (Tracy-Widom z-score ≤ +0.05, within noise). Group B yields four clear signal eigenvalues, validated against a difficulty-matched Bernoulli null (500 Monte Carlo draws), which produces at most one. So intra-model dimensionality ≤ 1 vs. cross-model dimensionality ≥ 4.
  • Why: temperature dilution. Temperature uniformly rescales logits — it does not differentiate domains or question types. Sampling variation is therefore unstructured noise, like wandering different corridors of the same library, while different models are entirely different libraries.
  • Locally committed, globally incoherent. When Qwen2.5-7B errs on a borderline question, 87.4% of resamples pick the same wrong answer (strong local commitment), yet these commitments show no systematic cross-question correlation — errors are isolated events, not symptom clusters, unlike diagnosable human knowledge gaps.
  • Engineering implications.
  • Self-consistency remains valid for selective prediction (deciding whether to answer) and flagging likely hallucinations.
  • It cannot reveal systematic capability weaknesses; model diversity is needed for that.
  • Model ensembles are not just variance reduction — they are epistemically distinct information sources, not "a more expensive version of 100 samples."
  • Philosophical note. Consistency may reflect "local inertia" (memorization) rather than understanding; genuine knowledge would show structured, predictable performance patterns across related problems — a structure the paper finds absent in sampled behavior of current LLMs.

References cited in the post

1. Ali, I. (2026). *Stochastic Sampling is Epistemically Shallow...* arXiv:2607.20464. EIML@ICML 2026. 2. Wang, X., et al. (2023). Self-Consistency Improves Chain of Thought Reasoning in Language Models. *ICLR*. 3. Lakshminarayanan, B., et al. (2017). Simple and Scalable Predictive Uncertainty Estimation using Deep Ensembles. *NeurIPS*. 4. Marčenko, V.A., & Pastur, L.A. (1967). Distribution of eigenvalues for some sets of random matrices. *Mathematics of the USSR-Sbornik*. 5. Farquhar, S., et al. (2024). Detecting Hallucinations in Large Language Models Using Semantic Entropy. *Nature*.

Tags

#llm#self-consistency#uncertainty-estimation#random-matrix-theory#model-ensembles#hallucination-detection#icml-2026#arxiv

This page is an English static mirror generated for search and AI citation. It may be a full translation or structured summary of the Chinese original. Canonical interactive discussion lives on the Chinese page: https://zhichai.net/topic/178447112