Yeonseung Chung

Seungone Kim, Dongkeun Yoon, Kiril Gashteovski et al.

With the advancement of AI capabilities, AI reviewers are beginning to be deployed in scientific peer review, yet their capability and credibility remain in question: many scientists simply view them as probabilistic systems without the expertise to evaluate research, while other researchers are more optimistic about their readiness without concrete evidence. Understanding what AI reviewers do well, where they fall short, and what challenges remain is essential. However, existing evaluations of AI reviewers have focused on whether their verdicts match human verdicts (e.g., score alignment, acceptance prediction), which is insufficient to characterize their capabilities and limits. In this paper, we close this gap through a large-scale expert annotation study, in which 45 domain scientists in Physical, Biological, and Health Sciences spent 469 hours rating 2,960 individual criticisms (each targeting one specific aspect of a paper) from human-written and AI-generated reviews of 82 Nature-family papers on correctness, significance, and sufficiency of evidence. On a composite of all three dimensions, a reviewing agent powered by GPT-5.2 scores above each paper's top-rated human reviewer (60.0% vs. 48.2%, p = 0.009), while all three AI reviewers (including Gemini 3.0 Pro and Claude Opus 4.5) exceed the lowest-rated human across every dimension. AI reviewers' accurate criticisms are also more often rated significant and well-evidenced, and surface a distinct 26% of issues no human raises. However, AI reviewers overlap far more than humans do (21% vs. 3% for cross-reviewer pairs), and exhibit 16 recurring weaknesses humans do not share, such as limited subfield knowledge, lack of long context management over multiple files, and overly critical stance on minor issues. Overall, our results position current AI reviewers as complements to, not substitutes for, human reviewers.

Dongwoo Kim, Yeonseung Chung, Alice Oh

In latent Dirichlet allocation (LDA), topics are multinomial distributions over the entire vocabulary. However, the vocabulary usually contains many words that are not relevant in forming the topics. We adopt a variable selection method widely used in statistical modeling as a dimension reduction tool and combine it with LDA. In this variable selection model for LDA (vsLDA), topics are multinomial distributions over a subset of the vocabulary, and by excluding words that are not informative for finding the latent topic structure of the corpus, vsLDA finds topics that are more robust and discriminative. We compare three models, vsLDA, LDA with symmetric priors, and LDA with asymmetric priors, on heldout likelihood, MCMC chain consistency, and document classification. The performance of vsLDA is better than symmetric LDA for likelihood and classification, better than asymmetric LDA for consistency and classification, and about the same in the other comparisons.