As large language models (LLMs) rapidly advance, evaluating their performance is critical. LLMs are trained on multilingual data, but their reasoning abilities are mainly evaluated using English datasets. Hence, robust evaluation frameworks are needed using high-quality non-English datasets, especially low-resource languages (LRLs). This study evaluates eight state-of-the-art (SOTA) LLMs on Latvian and Giriama using a Massive Multitask Language Understanding (MMLU) subset curated with native speakers for linguistic and cultural relevance. Giriama is benchmarked for the first time. Our evaluation shows that OpenAI's o1 model outperforms others across all languages, scoring 92.8% in English, 88.8% in Latvian, and 70.8% in Giriama on 0-shot tasks. Mistral-large (35.6%) and Llama-70B IT (41%) have weak performance, on both Latvian and Giriama. Our results underscore the need for localized benchmarks and human evaluations in advancing cultural AI contextualization.
In machine learning (ML), it is common to account for multiple objectives when, e.g., selecting a model to deploy. However, it is often unclear how one should compare, aggregate and, ultimately, trade-off these objectives, as they might be measured in different units or scales. For example, when deploying large language models (LLMs), we might not only care about their performance, but also their CO2 consumption. In this work, we investigate how objectives can be sensibly compared and aggregated to navigate their Pareto front. To do so, we propose to make incomparable objectives comparable via their CDFs, approximated by their relative rankings. This allows us to aggregate them while matching user-specific preferences, allowing practitioners to meaningfully navigate and search for models in the Pareto front. We demonstrate the potential impact of our methodology in diverse areas such as LLM selection, domain generalization, and AutoML benchmarking, where classical ways to aggregate and normalize objectives fail.
This paper introduces JuDGE (Judgment Document Generation Evaluation), a novel benchmark for evaluating the performance of judgment document generation in the Chinese legal system. We define the task as generating a complete legal judgment document from the given factual description of the case. To facilitate this benchmark, we construct a comprehensive dataset consisting of factual descriptions from real legal cases, paired with their corresponding full judgment documents, which serve as the ground truth for evaluating the quality of generated documents. This dataset is further augmented by two external legal corpora that provide additional legal knowledge for the task: one comprising statutes and regulations, and the other consisting of a large collection of past judgment documents. In collaboration with legal professionals, we establish a comprehensive automated evaluation framework to assess the quality of generated judgment documents across various dimensions. We evaluate various baseline approaches, including few-shot in-context learning, fine-tuning, and a multi-source retrieval-augmented generation (RAG) approach, using both general and legal-domain LLMs. The experimental results demonstrate that, while RAG approaches can effectively improve performance in this task, there is still substantial room for further improvement. All the codes and datasets are available at: https://github.com/oneal2000/JuDGE.
Large language models (LLMs) have been one of the most important discoveries in machine learning in recent years. LLM-based artificial intelligence (AI) assistants, such as ChatGPT, have consistently attracted the attention from researchers, investors, and the general public, driving the rapid growth of this industry. With the frequent introduction of new LLMs to the market, it becomes increasingly difficult to differentiate between them, creating a demand for new LLM comparison methods. In this research, the Consistency-focused Similarity Comparison Framework (ConSCompF) for generative large language models is proposed. It compares texts generated by two LLMs and produces a similarity score, indicating the overall degree of similarity between their responses. The main advantage of this framework is that it can operate on a small number of unlabeled data, such as chatbot instruction prompts, and does not require LLM developers to disclose any information about their product. To evaluate the efficacy of ConSCompF, two experiments aimed at identifying similarities between multiple LLMs are conducted. Additionally, these experiments examine the correlation between the similarity scores generated by ConSCompF and the differences in the outputs produced by other benchmarking techniques, such as ROUGE-L. Finally, a series of few-shot LLM comparison experiments is conducted to evaluate the performance of ConSCompF in a few-shot LLM comparison scenario. The proposed framework can be used for calculating similarity matrices of multiple LLMs, which can be effectively visualized using principal component analysis (PCA). The ConSCompF output may provide useful insights into data that might have been used during LLM training and help detect possible investment fraud attempts.
We present Stable Virtual Camera (Seva), a generalist diffusion model that creates novel views of a scene, given any number of input views and target cameras. Existing works struggle to generate either large viewpoint changes or temporally smooth samples, while relying on specific task configurations. Our approach overcomes these limitations through simple model design, optimized training recipe, and flexible sampling strategy that generalize across view synthesis tasks at test time. As a result, our samples maintain high consistency without requiring additional 3D representation-based distillation, thus streamlining view synthesis in the wild. Furthermore, we show that our method can generate high-quality videos lasting up to half a minute with seamless loop closure. Extensive benchmarking demonstrates that Seva outperforms existing methods across different datasets and settings.
Vision-Language Models (VLMs) have recently witnessed significant progress in visual comprehension. As the permitting length of image context grows, VLMs can now comprehend a broader range of views and spaces. Current benchmarks provide insightful analysis of VLMs in tasks involving complex visual instructions following, multi-image understanding and spatial reasoning. However, they usually focus on spatially irrelevant images or discrete images captured from varied viewpoints. The compositional characteristic of images captured from a static viewpoint remains underestimated. We term this characteristic as Continuous Space Perception. When observing a scene from a static viewpoint while shifting orientations, it produces a series of spatially continuous images, enabling the reconstruction of the entire space. In this paper, we present CoSpace, a multi-image visual understanding benchmark designed to assess the Continuous Space perception ability for VLMs. CoSpace contains 2,918 images and 1,626 question-answer pairs, covering seven types of tasks. We conduct evaluation across 19 proprietary and open-source VLMs. Results reveal that there exist pitfalls on the continuous space perception ability for most of the evaluated models, including proprietary ones. Interestingly, we find that the main discrepancy between open-source and proprietary models lies not in accuracy but in the consistency of responses. We believe that enhancing the ability of continuous space perception is essential for VLMs to perform effectively in real-world tasks and encourage further research to advance this capability.
Deep learning (DL) and machine learning (ML) models have shown promise in drug response prediction (DRP), yet their ability to generalize across datasets remains an open question, raising concerns about their real-world applicability. Due to the lack of standardized benchmarking approaches, model evaluations and comparisons often rely on inconsistent datasets and evaluation criteria, making it difficult to assess true predictive capabilities. In this work, we introduce a benchmarking framework for evaluating cross-dataset prediction generalization in DRP models. Our framework incorporates five publicly available drug screening datasets, six standardized DRP models, and a scalable workflow for systematic evaluation. To assess model generalization, we introduce a set of evaluation metrics that quantify both absolute performance (e.g., predictive accuracy across datasets) and relative performance (e.g., performance drop compared to within-dataset results), enabling a more comprehensive assessment of model transferability. Our results reveal substantial performance drops when models are tested on unseen datasets, underscoring the importance of rigorous generalization assessments. While several models demonstrate relatively strong cross-dataset generalization, no single model consistently outperforms across all datasets. Furthermore, we identify CTRPv2 as the most effective source dataset for training, yielding higher generalization scores across target datasets. By sharing this standardized evaluation framework with the community, our study aims to establish a rigorous foundation for model comparison, and accelerate the development of robust DRP models for real-world applications.
Accurate segmentation is essential for effective treatment planning and disease monitoring. Existing medical image segmentation methods predominantly rely on uni-modal visual inputs, such as images or videos, requiring labor-intensive manual annotations. Additionally, medical imaging techniques capture multiple intertwined organs within a single scan, further complicating segmentation accuracy. To address these challenges, MedSAM, a large-scale medical segmentation model based on the Segment Anything Model (SAM), was developed to enhance segmentation accuracy by integrating image features with user-provided prompts. While MedSAM has demonstrated strong performance across various medical segmentation tasks, it primarily relies on geometric prompts (e.g., points and bounding boxes) and lacks support for text-based prompts, which could help specify subtle or ambiguous anatomical structures. To overcome these limitations, we propose the Organ-aware Multi-scale Text-guided Medical Image Segmentation Model (OMT-SAM) for multi-organ segmentation. Our approach introduces CLIP encoders as a novel image-text prompt encoder, operating with the geometric prompt encoder to provide informative contextual guidance. We pair descriptive textual prompts with corresponding images, processing them through pre-trained CLIP encoders and a cross-attention mechanism to generate fused image-text embeddings. Additionally, we extract multi-scale visual features from MedSAM, capturing fine-grained anatomical details at different levels of granularity. We evaluate OMT-SAM on the FLARE 2021 dataset, benchmarking its performance against existing segmentation methods. Empirical results demonstrate that OMT-SAM achieves a mean Dice Similarity Coefficient of 0.937, outperforming MedSAM (0.893) and other segmentation models, highlighting its superior capability in handling complex medical image segmentation tasks.
Understanding fine-grained temporal dynamics is crucial in egocentric videos, where continuous streams capture frequent, close-up interactions with objects. In this work, we bring to light that current egocentric video question-answering datasets often include questions that can be answered using only few frames or commonsense reasoning, without being necessarily grounded in the actual video. Our analysis shows that state-of-the-art Multi-Modal Large Language Models (MLLMs) on these benchmarks achieve remarkably high performance using just text or a single frame as input. To address these limitations, we introduce EgoTempo, a dataset specifically designed to evaluate temporal understanding in the egocentric domain. EgoTempo emphasizes tasks that require integrating information across the entire video, ensuring that models would need to rely on temporal patterns rather than static cues or pre-existing knowledge. Extensive experiments on EgoTempo show that current MLLMs still fall short in temporal reasoning on egocentric videos, and thus we hope EgoTempo will catalyze new research in the field and inspire models that better capture the complexity of temporal dynamics. Dataset and code are available at https://github.com/google-research-datasets/egotempo.git.
Scientific research demands sophisticated reasoning over multimodal data, a challenge especially prevalent in biology. Despite recent advances in multimodal large language models (MLLMs) for AI-assisted research, existing multimodal reasoning benchmarks only target up to college-level difficulty, while research-level benchmarks emphasize lower-level perception, falling short of the complex multimodal reasoning needed for scientific discovery. To bridge this gap, we introduce MicroVQA, a visual-question answering (VQA) benchmark designed to assess three reasoning capabilities vital in research workflows: expert image understanding, hypothesis generation, and experiment proposal. MicroVQA consists of 1,042 multiple-choice questions (MCQs) curated by biology experts across diverse microscopy modalities, ensuring VQA samples represent real scientific practice. In constructing the benchmark, we find that standard MCQ generation methods induce language shortcuts, motivating a new two-stage pipeline: an optimized LLM prompt structures question-answer pairs into MCQs; then, an agent-based `RefineBot' updates them to remove shortcuts. Benchmarking on state-of-the-art MLLMs reveal a peak performance of 53\%; models with smaller LLMs only slightly underperform top models, suggesting that language-based reasoning is less challenging than multimodal reasoning; and tuning with scientific articles enhances performance. Expert analysis of chain-of-thought responses shows that perception errors are the most frequent, followed by knowledge errors and then overgeneralization errors. These insights highlight the challenges in multimodal scientific reasoning, showing MicroVQA is a valuable resource advancing AI-driven biomedical research. MicroVQA is available at https://huggingface.co/datasets/jmhb/microvqa, and project page at https://jmhb0.github.io/microvqa.
Background: Data driven stratification of patients into biologically informed subtypes holds promise for precision neuropsychiatry, yet neuroimaging-based clustering methods often fail to generalize across cohorts. While algorithmic innovations have focused on model complexity, the role of underlying dataset characteristics remains underexplored. We hypothesized that cluster separation, size imbalance, noise, and the direction and magnitude of disease-related effects in the input data critically determine both within-algorithm accuracy and reproducibility. Methods: We evaluated 4 widely used stratification algorithms, HYDRA, SuStaIn, SmileGAN, and SurrealGAN, on a suite of synthetic brain-morphometry cohorts derived from the Human Connectome Project Young Adult dataset. Three global transformation patterns were applied to 600 pseudo-patients against 508 controls, followed by 4 within-dataset variations varying cluster count (k=2-6), overlap, and effect magnitude. Algorithm performance was quantified by accuracy in recovering the known ground-truth clusters. Results: Across 122 synthetic scenarios, data complexity consistently outweighed algorithm choice in predicting stratification success. Well-separated clusters yielded high accuracy for all methods, whereas overlapping, unequal-sized, or subtle effects reduced accuracy by up to 50%. SuStaIn could not scale beyond 17 features, HYDRA's accuracy varied unpredictably with data heterogeneity. SmileGAN and SurrealGAN maintained robust pattern detection but did not assign discrete cluster labels to individuals. Conclusions: The study results demonstrate the impact of statistical properties of input data across algorithms and highlight the need for using realistic dataset distributions when new algorithms are being developed and suggest greater focus on data-centric strategies that actively shape and standardize the input distributions.
Species distribution models (SDMs) are necessary for measuring and predicting occurrences and habitat suitability of species and their relationship with environmental factors. We introduce a novel presence-only SDM with graph neural networks (GNN). In our model, species and locations are treated as two distinct node sets, and the learning task is predicting detection records as the edges that connect locations to species. Using GNN for SDM allows us to model fine-grained interactions between species and the environment. We evaluate the potential of this methodology on the six-region dataset compiled by National Center for Ecological Analysis and Synthesis (NCEAS) for benchmarking SDMs. For each of the regions, the heterogeneous GNN model is comparable to or outperforms previously-benchmarked single-species SDMs as well as a feed-forward neural network baseline model.
Clinically acquired brain MRIs and radiology reports are valuable but underutilized resources due to the challenges of manual analysis and data heterogeneity. We developed fine-tuned language models (LMs) to classify brain MRI reports as normal (reports with limited pathology) or abnormal, fine-tuning BERT, BioBERT, ClinicalBERT, and RadBERT on 44,661 reports. We also explored the reasoning capabilities of a leading LM, Gemini 1.5-Pro, for normal report categorization. Automated image processing and modeling generated brain growth charts from LM-classified normal scans, comparing them to human-derived charts. Fine-tuned LMs achieved high classification performance (F1-Score >97%), with unbalanced training mitigating class imbalance. Performance was robust on out-of-distribution data, with full text outperforming summary (impression) sections. Gemini 1.5-Pro showed a promising categorization performance, especially with clinical inference. LM-derived brain growth charts were nearly identical to human-annotated charts (r = 0.99, p < 2.2e-16). Our LMs offer scalable analysis of radiology reports, enabling automated classification of brain MRIs in large datasets. One application is automated generation of brain growth charts for benchmarking quantitative image features. Further research is needed to address data heterogeneity and optimize LM reasoning.
The integration of tools has extended the capabilities of language models (LMs) beyond vanilla text generation to versatile scenarios. However, tool-augmented language models (TaLMs) often assume 'perfect' information access and tool availability, which may not hold in the real world. To systematically study TaLMs' imperfections, we introduce the FAIL-TALMS benchmark, featuring two major failures: under-specified user queries and non-available tools. FAIL-TALMS contains 1,749 examples using 906 tools across 21 categories, including single- and multi-tool usage. We evaluate top-performing proprietary and open-source models, and find all current models except for Claude struggle to recognize missing tools or information. Further, to study possible mitigation of the failures, we enable real-time human interaction, named the Ask-and-Help (AAH) method, to provide missing information or replace non-functional tools. While AAH can help models solve tasks more correctly when queries are under-specified, it brings minimal benefit when complex tools are broken.
Vision-and-Language Navigation (VLN) systems often focus on either discrete (panoramic) or continuous (free-motion) paradigms alone, overlooking the complexities of human-populated, dynamic environments. We introduce a unified Human-Aware VLN (HA-VLN) benchmark that merges these paradigms under explicit social-awareness constraints. Our contributions include: 1. A standardized task definition that balances discrete-continuous navigation with personal-space requirements; 2. An enhanced human motion dataset (HAPS 2.0) and upgraded simulators capturing realistic multi-human interactions, outdoor contexts, and refined motion-language alignment; 3. Extensive benchmarking on 16,844 human-centric instructions, revealing how multi-human dynamics and partial observability pose substantial challenges for leading VLN agents; 4. Real-world robot tests validating sim-to-real transfer in crowded indoor spaces; and 5. A public leaderboard supporting transparent comparisons across discrete and continuous tasks. Empirical results show improved navigation success and fewer collisions when social context is integrated, underscoring the need for human-centric design. By releasing all datasets, simulators, agent code, and evaluation tools, we aim to advance safer, more capable, and socially responsible VLN research.
Large Language Models (LLMs) have revolutionized code generation, achieving exceptional results on various established benchmarking frameworks. However, concerns about data contamination - where benchmark data inadvertently leaks into pre-training or fine-tuning datasets - raise questions about the validity of these evaluations. While this issue is known, limiting the industrial adoption of LLM-driven software engineering, hardware coding has received little to no attention regarding these risks. For the first time, we analyze state-of-the-art (SOTA) evaluation frameworks for Verilog code generation (VerilogEval and RTLLM), using established methods for contamination detection (CCD and Min-K% Prob). We cover SOTA commercial and open-source LLMs (CodeGen2.5, Minitron 4b, Mistral 7b, phi-4 mini, LLaMA-{1,2,3.1}, GPT-{2,3.5,4o}, Deepseek-Coder, and CodeQwen 1.5), in baseline and fine-tuned models (RTLCoder and Verigen). Our study confirms that data contamination is a critical concern. We explore mitigations and the resulting trade-offs for code quality vs fairness (i.e., reducing contamination toward unbiased benchmarking).
In recent years, the rapid advancement of Artificial Intelligence (AI) technologies, particularly Large Language Models (LLMs), has revolutionized the paradigm of scientific discovery, establishing AI-for-Science (AI4Science) as a dynamic and evolving field. However, there is still a lack of an effective framework for the overall assessment of AI4Science, particularly from a holistic perspective on data quality and model capability. Therefore, in this study, we propose SciHorizon, a comprehensive assessment framework designed to benchmark the readiness of AI4Science from both scientific data and LLM perspectives. First, we introduce a generalizable framework for assessing AI-ready scientific data, encompassing four key dimensions: Quality, FAIRness, Explainability, and Compliance-which are subdivided into 15 sub-dimensions. Drawing on data resource papers published between 2018 and 2023 in peer-reviewed journals, we present recommendation lists of AI-ready datasets for Earth, Life, and Materials Sciences, making a novel and original contribution to the field. Concurrently, to assess the capabilities of LLMs across multiple scientific disciplines, we establish 16 assessment dimensions based on five core indicators Knowledge, Understanding, Reasoning, Multimodality, and Values spanning Mathematics, Physics, Chemistry, Life Sciences, and Earth and Space Sciences. Using the developed benchmark datasets, we have conducted a comprehensive evaluation of over 50 representative open-source and closed source LLMs. All the results are publicly available and can be accessed online at www.scihorizon.cn/en.
An interesting class of commonsense reasoning problems arises when people are faced with natural disasters. To investigate this topic, we present \textsf{RESPONSE}, a human-curated dataset containing 1789 annotated instances featuring 6037 sets of questions designed to assess LLMs' commonsense reasoning in disaster situations across different time frames. The dataset includes problem descriptions, missing resources, time-sensitive solutions, and their justifications, with a subset validated by environmental engineers. Through both automatic metrics and human evaluation, we compare LLM-generated recommendations against human responses. Our findings show that even state-of-the-art models like GPT-4 achieve only 37\% human-evaluated correctness for immediate response actions, highlighting significant room for improvement in LLMs' ability for commonsense reasoning in crises.
NLP research has increasingly focused on subjective tasks such as emotion analysis. However, existing emotion benchmarks suffer from two major shortcomings: (1) they largely rely on keyword-based emotion recognition, overlooking crucial cultural dimensions required for deeper emotion understanding, and (2) many are created by translating English-annotated data into other languages, leading to potentially unreliable evaluation. To address these issues, we introduce Cultural Lenses on Emotion (CuLEmo), the first benchmark designed to evaluate culture-aware emotion prediction across six languages: Amharic, Arabic, English, German, Hindi, and Spanish. CuLEmo comprises 400 crafted questions per language, each requiring nuanced cultural reasoning and understanding. We use this benchmark to evaluate several state-of-the-art LLMs on culture-aware emotion prediction and sentiment analysis tasks. Our findings reveal that (1) emotion conceptualizations vary significantly across languages and cultures, (2) LLMs performance likewise varies by language and cultural context, and (3) prompting in English with explicit country context often outperforms in-language prompts for culture-aware emotion and sentiment understanding. We hope this benchmark guides future research toward developing more culturally aligned NLP systems.
Visual reasoning is central to human cognition, enabling individuals to interpret and abstractly understand their environment. Although recent Multimodal Large Language Models (MLLMs) have demonstrated impressive performance across language and vision-language tasks, existing benchmarks primarily measure recognition-based skills and inadequately assess true visual reasoning capabilities. To bridge this critical gap, we introduce VERIFY, a benchmark explicitly designed to isolate and rigorously evaluate the visual reasoning capabilities of state-of-the-art MLLMs. VERIFY compels models to reason primarily from visual information, providing minimal textual context to reduce reliance on domain-specific knowledge and linguistic biases. Each problem is accompanied by a human-annotated reasoning path, making it the first to provide in-depth evaluation of model decision-making processes. Additionally, we propose novel metrics that assess visual reasoning fidelity beyond mere accuracy, highlighting critical imbalances in current model reasoning patterns. Our comprehensive benchmarking of leading MLLMs uncovers significant limitations, underscoring the need for a balanced and holistic approach to both perception and reasoning. For more teaser and testing, visit our project page (https://verify-eqh.pages.dev/).