Large language models (LLMs) are increasingly used as automated evaluators of AI systems, including in high-stakes applications. In this role, LLMs are used to generate judgments about the quality, appropriateness, or even safety of model outputs. This approach is motivated by practical constraints. Expert human ratings are costly and difficult to scale, whereas LLM ratings can be produced quickly at low cost. However, current approaches to deploying LLM evaluators are ad hoc, typically limited to reporting agreement metrics between human and LLM judges as a justification for substitution of human ratings, and lack a formal basis for study design. This paper (1) shifts the role of the LLM judge from substitutive to auxiliary, and (2) formulates the LLM-as-a-judge paradigm as one of augmenting human evaluation through a two-stage sampling design, where LLM evaluations are measured for all observations at the first stage and human ratings are partially observed for a subsample at the second stage. We propose to use a doubly robust estimator from the missing data literature, which takes advantage of the robustness property against the prediction model, since the missingness model is known by design. Using the asymptotic variance of this estimator, we propose how sample sizes of human and LLM ratings can be determined to achieve a targeted level of power. We also show that a study can be efficiently designed by allocating more human ratings for types of evaluations where the predictability of LLM ratings is not high. To the best of our knowledge, there is very little guidance on how much human oversight should be retained when validating benchmarks.

Evaluation of large language models (LLMs) is increasingly critical, yet standard benchmarking methods rely on average accuracy, overlooking both the inherent stochasticity of LLM outputs and the heterogeneity of benchmark items. Item Response Theory (IRT) offers a principled framework for modeling latent model abilities and item characteristics, but conventional methods are computationally expensive and numerically unstable, limiting large-scale implementations. To address these challenges, we propose an interpretable and scalable framework for LLM evaluation based on the majorization-minimization principle. Our approach reformulates the problem as a sequence of constrained matrix factorization subproblems, enabling stable and efficient parameter estimation with theoretical guarantees for identifiability and convergence. Experiments on synthetic and real-world datasets, including MATH-500 and six Open LLM Leaderboard benchmarks, demonstrate that our method achieves superior scalability and interpretability. It delivers orders-of-magnitude speedups over competing methods while maintaining comparable or even higher estimation accuracy. Our results align with established scaling laws and offer insights into item difficulty and discrimination, informing more principled benchmark design.

Evaluating large language models (LLMs) is challenging. Traditional ground-truth-based benchmarks fail to capture the comprehensiveness and nuance of real-world queries, while LLM-as-judge benchmarks suffer from grading biases and limited query quantity. Both of them may also become contaminated over time. User-facing evaluation, such as Chatbot Arena, provides reliable signals but is costly and slow. In this work, we propose MixEval, a new paradigm for establishing efficient, gold-standard LLM evaluation by strategically mixing off-the-shelf benchmarks.

We find that the distribution shift settings in previous studies commonly lack adequate challenges, hindering the accurate evaluation of OOD robustness. To address these issues, we propose a benchmark construction protocol that ensures clear differentiation and challenging distribution shifts.

LLM leaderboards often rely on single stochastic runs, but how many repetitions are required for reliable conclusions remains unclear. We re-evaluate eight state-of-the-art models on the AI4Math Benchmark with three independent runs per setting. Using mixed-effects logistic regression, domain-level marginal means, rank-instability analysis, and run-to-run reliability, we assessed the value of additional repetitions. Our findings shows that Single-run leaderboards are brittle: 10/12 slices (83\%) invert at least one pairwise rank relative to the three-run majority, despite a zero sign-flip rate for pairwise significance and moderate overall interclass correlation. Averaging runs yields modest SE shrinkage ($\sim$5\% from one to three) but large ranking gains; two runs remove $\sim$83\% of single-run inversions. We provide cost-aware guidance for practitioners: treat evaluation as an experiment, report uncertainty, and use $\geq 2$ repetitions under stochastic decoding. These practices improve robustness while remaining feasible for small teams and help align model comparisons with real-world reliability.

Large Language Models (LLMs) are accelerating scientific idea generation, but rigorously evaluating these numerous, often superficial, AI-generated propositions for novelty and factual accuracy is a critical bottleneck; manual verification is too slow. Existing validation methods are inadequate: LLMs as standalone verifiers may hallucinate and lack domain knowledge (our findings show 60% unawareness of relevant papers in specific domains), while traditional citation networks lack explicit causality and narrative surveys are unstructured. This underscores a core challenge: the absence of structured, verifiable, and causally-linked historical data of scientific evolution.To address this,we introduce \textbf{THE-Tree} (\textbf{T}echnology \textbf{H}istory \textbf{E}volution Tree), a computational framework that constructs such domain-specific evolution trees from scientific literature. THE-Tree employs a search algorithm to explore evolutionary paths. During its node expansion, it utilizes a novel "Think-Verbalize-Cite-Verify" process: an LLM proposes potential advancements and cites supporting literature. Critically, each proposed evolutionary link is then validated for logical coherence and evidential support by a recovered natural language inference mechanism that interrogates the cited literature, ensuring that each step is grounded. We construct and validate 88 THE-Trees across diverse domains and release a benchmark dataset including up to 71k fact verifications covering 27k papers to foster further research. Experiments demonstrate that i) in graph completion, our THE-Tree improves hit@1 by 8% to 14% across multiple models compared to traditional citation networks; ii) for predicting future scientific developments, it improves hit@1 metric by nearly 10%; and iii) when combined with other methods, it boosts the performance of evaluating important scientific papers by almost 100%.

People get informed of a daily task plan through diverse media involving both texts and images. However, most prior research only focuses on LLM's capability of textual plan generation. The potential of large-scale models in providing text-image plans remains understudied. Generating high-quality text-image plans faces two main challenges: ensuring consistent alignment between two modalities and keeping coherence among visual steps. To address these challenges, we propose a novel framework that generates and refines text-image plans step-by-step. At each iteration, our framework (1) drafts the next textual step based on the prediction history; (2) edits the last visual step to obtain the next one; (3) extracts PDDL-like visual information; and (4) refines the draft with the extracted visual information. The textual and visual step produced in stage (4) and (2) will then serve as inputs for the next iteration. Our approach offers a plug-and-play improvement to various backbone models, such as Mistral-7B, Gemini-1.5, and GPT-4o. To evaluate the effectiveness of our approach, we collect a new benchmark consisting of 1,100 tasks and their text-image pair solutions covering 11 daily topics. We also design and validate a new set of metrics to evaluate the multimodal consistency and coherence in text-image plans. Extensive experiment results show the effectiveness of our approach on a range of backbone models against competitive baselines. Our code and data are available at https://github.com/psunlpgroup/MPlanner.

Online learning has experienced rapid growth due to its flexibility and accessibility. Personalization, adapted to the needs of individual learners, is crucial for enhancing the learning experience, particularly in online settings. A key aspect of personalization is providing learners with answers customized to their specific questions. This paper therefore explores the potential of Large Language Models ( LLMs) to generate personalized answers to learners' questions, thereby enhancing engagement and reducing the workload on educators. To evaluate the effectiveness of LLMs in this context, we conducted a comprehensive study using the StackExchange platform in two distinct areas: language learning and programming. We developed a framework and a dataset for validating automatically generated personalized answers. Subsequently, we generated personalized answers using different strategies, including 0-shot, 1-shot, and few-shot scenarios. The generated answers were evaluated using three methods: 1. BERTScore, 2. LLM evaluation, and 3. human evaluation. Our findings indicated that providing LLMs with examples of desired answers (from the learner or similar learners) can significantly enhance the LLMs' ability to tailor responses to individual learners' needs.

This research introduces ScoreRAG, an approach to enhance the quality of automated news generation. Despite advancements in Natural Language Processing and large language models, current news generation methods often struggle with hallucinations, factual inconsistencies, and lack of domain-specific expertise when producing news articles. ScoreRAG addresses these challenges through a multi-stage framework combining retrieval-augmented generation, consistency relevance evaluation, and structured summarization. The system first retrieves relevant news documents from a vector database, maps them to complete news items, and assigns consistency relevance scores based on large language model evaluations. These documents are then reranked according to relevance, with low-quality items filtered out. The framework proceeds to generate graded summaries based on relevance scores, which guide the large language model in producing complete news articles following professional journalistic standards. Through this methodical approach, ScoreRAG aims to significantly improve the accuracy, coherence, informativeness, and professionalism of generated news articles while maintaining stability and consistency throughout the generation process. The code and demo are available at: https://github.com/peiyun2260/ScoreRAG.

Evaluating large language models (LLMs) is challenging. Traditional ground-truth- based benchmarks fail to capture the comprehensiveness and nuance of real-world queries, while LLM-as-judge benchmarks suffer from grading biases and limited query quantity. Both of them may also become contaminated over time. User- facing evaluation, such as Chatbot Arena, provides reliable signals but is costly and slow. In this work, we propose MixEval, a new paradigm for establishing efficient, gold-standard LLM evaluation by strategically mixing off-the-shelf bench- marks.