acceptodds
Under review as a conference paper at ICLR 2027

PROBABILISTIC EVALUATION OF LARGE LANGUAGE MODELS WITH AMORTIZED BAYESIAN RELIABILITY ESTIMATION

Abstract

Large language model (LLM) evaluation increasingly relies on automated performance metrics to compare models across tasks, yet the reliability and uncertainty of these evaluations remain insufficiently characterized. Conventional evaluation metrics typically provide point estimates of model performance, making it difficult to determine whether an observed performance difference reflects a meaningful improvement or uncertainty in the evaluation process. In this work, we propose a Bayesian inference-based performance metric for quantifying the reliability of LLM evaluation. The proposed approach models evaluation outcomes probabilistically, enabling performance estimates to be accompanied by uncertainty information and providing a principled basis for assessing the confidence and reliability of model comparisons. We evaluate the proposed metric across multiple language models and benchmarks, including PolyGemma, BERT, and Polyglot, and compare its behavior against established evaluation approaches. Our experimental analysis demonstrates how Bayesian inference can expose uncertainty that is otherwise hidden by conventional point-based metrics and provide a more informative characterization of model performance. The results highlight the importance of accounting for uncertainty when ranking or comparing LLMs and suggest that reliable evaluation should consider not only the magnitude of measured performance but also the statistical confidence associated with that measurement. The proposed framework provides a general and interpretable approach for uncertainty-aware evaluation of LLMs and can be applied to systematic comparisons across models, datasets, and evaluation settings.We formulate LLM evaluation as Bayesian inference over a continuous latent semantic-quality variable, using a Gaussian prior to represent prior beliefs and uncertainty over graded semantic relatedness.

Primary area:
other topics in machine learning (i.e., none of the above)
Keywords:
Large Language Models (LLMs), Bayesian Inference, Uncertainty Quantification, LLM Evaluation, Performance Reliability, Probabilistic Evaluation, Model Comparison, Robust Evaluation.

Reviews

Among:
Shape:
1234this paper, Bottom 1%
Figure 1. Mean review rating of the reviewed papers at ICLR 2027. This paper averages 1.00 over 3 reviews, above 0% of the others.
Reviewer_3uXBOpenReview ↗
Rating
1
Confidence
3
Informativeness
1
Presentation
1
Soundness
1
Read review
Paper summary

LLM leaderboards usually report point estimates without uncertainty, and this paper responds with Amortized Bayesian Reliability Estimation. A model's per-item benchmark scores are treated as i.i.d. draws given a latent performance parameter θ, Bernoulli for pass/fail scoring and a mean-parameterized Beta for continuous scores, with a Beta prior on θ. A permutation-invariant network trained by neural posterior estimation on simulated or historical evaluation tasks maps a set of scores to a Beta approximation of the posterior in one forward pass. The posterior mean is reported with credible intervals, and models are compared through the probability of superiority, a ROPE probability and rank distributions; a hierarchical variant pools across benchmark subgroups. The claimed contributions are this Bayesian metric, reliability-aware model comparison, and an empirical evaluation on PolyGemma, BERT and Polyglot. The abstract and Section 2.6 describe a different formulation, with a Gaussian prior over a latent semantic-quality variable and an LLM judging semantic relatedness. On validation, Sections 4.12–4.14 set out a protocol, ablations and an experimental setup, but no experimental results are reported.

Critical strengths

Uncertainty in LLM evaluation is a real problem worth solving. Models are routinely ranked on point estimates from benchmarks of a few hundred items, and a practical way to report how far those numbers can be trusted would improve how comparisons are made. The reporting design in Section 4 is sensible: it keeps the estimate separate from its uncertainty, plans to check calibration with coverage curves at several nominal levels, and fixes the ROPE margin before the data are seen. This carries limited weight in my assessment, because none of it is validated in the paper and these reporting practices already exist in the literature on Bayesian evaluation.

Critical weaknesses

The paper does not report any experiments. There are no result tables or figures, and the experimental setup in Section 4.14 still contains notes such as "[Add exact model variants/sizes and checkpoints ...]" and a request to list the benchmark datasets. It looks as though the experiments were never fixed, let alone run. Even so, the abstract says the metric was evaluated across several models and benchmarks and that "our experimental analysis demonstrates" uncertainty that point estimates hide, and a comprehensive empirical evaluation is listed as the third contribution. Other parts of the paper are unfinished as well: Section 1.1 is the ICLR template text, the reference list is empty, and Section 2 refers to the works it discusses only as "Paper 1" to "Paper 6". This on its own decides my recommendation, because none of the paper's claims can be checked.

Setting the missing experiments aside, I could not work out what the method actually is, and the version described in most detail has a flaw at its core. The abstract and Section 2.6 describe an LLM judging the semantic relationship between prompt, response and reference, with a Gaussian prior on a continuous latent quality variable. Section 4 instead trains an amortized network that outputs a Beta posterior, using a uniform prior by default, and Algorithm 1 uses closed-form updates or MCMC with no network at all. In the amortized version, the set encoder in Eq. 11 averages the per-item embeddings, so it never sees the number of items N. For binary scores its input is just the accuracy. A model that gets 7 of 10 items right and one that gets 700 of 1000 right therefore receive the same posterior, although the exact posterior is about ten times narrower in the second case. Measuring how much an evaluation set tells us about performance is the point of the paper, so this undermines the main contribution. I also don't see why amortization is needed, since the exact posterior is closed-form for binary scores and a cheap one- or two-dimensional integral for the Beta likelihood.

Questions for authors

A few clarifications could change my assessment. The most important is whether any experiments were run. The abstract and the contribution list describe results, but the paper contains none, so if results exist, please report the models, benchmarks and numbers. I would also like to know which of the three formulations is the actual contribution: the Gaussian latent-quality model with an LLM judge in the abstract and Section 2.6, the amortized Beta posterior of Section 4, or the conjugate and MCMC procedure of Algorithm 1. How does the encoder in Eq. 11 take the number of items N into account? As written it averages over items, and I do not see how the posterior can then narrow as N grows. Finally, the comparisons in Sections 4.9 and 4.10 treat the models' posteriors as independent. How do they handle the standard case where all models are scored on the same items?

Reviewer_44FEOpenReview ↗
Rating
1
Confidence
3
Informativeness
1
Presentation
1
Soundness
1
Read review
Paper summary

The paper proposes Amortized Bayesian Reliability Estimation (ABRE) to address the limitations of traditional evaluation metrics, which provide only point estimates of model performance and make it difficult to determine whether observed performance differences reflect meaningful improvements or uncertainty.

However, the paper version currently available for reviewers to download appears to be an incomplete draft. It exceeds the page limit (9 pages of main body) and lacks experimental results and analysis, references, and other essential components. As a result, it is difficult to conduct a substantive review.

Critical strengths
  • The paper addresses a meaningful motivation. Distinguishing whether observed differences in evaluation results reflect substantive performance differences or evaluation uncertainty is indeed a major challenge in current LLM evaluation. Addressing this issue would be important for evaluating LLMs more reliably and improving the efficiency of LLM development.
Critical weaknesses
  • The currently submitted version appears to be an incomplete draft with numerous serious issues, making a substantive review impossible. For example, the main text exceeds the nine-page limit, and I do not understand why the paper was not desk-rejected. Section 1 (Introduction) ends after presenting the background and contributions, while Section 1.1 remains in the default LaTeX template state. Section 2 (Related Work) occupies a substantial portion of the paper (pages 3–6), yet the References section is empty, and the appendix also remains in the default LaTeX template state. Section 4 contains numerous subsections that intermingle methodology and experiments, but no experimental results are provided.
Questions for authors

None

Reviewer_5xUCOpenReview ↗
Rating
1
Confidence
4
Informativeness
1
Presentation
1
Soundness
1
Read review
Paper summary

The paper attempts to introduce a Bayesian approach for estimating a reliability parameter that characterizes LLM performance. However there are no novel contributions because the paper is clearly incomplete and fails to adhere to ICLR guidelines: it has no references, no results (and no figures), no appendices, and the main text runs over the 9 page limit.

Critical strengths

None.

Critical weaknesses

As mentioned above, this is an incomplete submission that does not adhere to paper guidelines, so does not merit further evaluation.

Questions for authors

None.

From OpenReview, CC BY 4.0. Authors as signed there; shown unedited.

Related papers

Loading the map…

Discussion (0)

Sign in to comment.