As AI systems increasingly evaluate other AI outputs, understanding their assessment behavior becomes crucial for preventing cascading biases. This study analyzes vision-language descriptions generated by NVIDIA's Describe Anything Model and evaluated by three GPT variants (GPT-4o, GPT-4o-mini, GPT-5) to uncover distinct "evaluation personalities" the underlying assessment strategies and biases each model demonstrates. GPT-4o-mini exhibits systematic consistency with minimal variance, GPT-4o excels at error detection, while GPT-5 shows extreme conservatism with high variability. Controlled experiments using Gemini 2.5 Pro as an independent question generator validate that these personalities are inherent model properties rather than artifacts. Cross-family analysis through semantic similarity of generated questions reveals significant divergence: GPT models cluster together with high similarity while Gemini exhibits markedly different evaluation strategies. All GPT models demonstrate a consistent 2:1 bias favoring negative assessment over positive confirmation, though this pattern appears family-specific rather than universal across AI architectures. These findings suggest that evaluation competence does not scale with general capability and that robust AI assessment requires diverse architectural perspectives.

The rapid scaling of Large Language Models (LLMs) has pushed training workloads far beyond the limits of single-node analysis, demanding a deeper understanding of how these models behave across large-scale, multi-GPU systems. In this paper, we present a comprehensive characterization of LLM training across diverse real-world workloads and hardware platforms, including NVIDIA H100/H200 and AMD MI250 GPUs. We analyze dense and sparse models under various parallelism strategies -- tensor, pipeline, data, and expert -- and evaluate their effects on hardware utilization, power consumption, and thermal behavior. We further evaluate the effectiveness of optimizations such as activation recomputation and compute-communication overlap. Our findings show that performance is not determined solely by scaling hardware capacity. Scale-up systems with fewer, higher-memory GPUs can outperform scale-out systems in communication-bound regimes, but only under carefully tuned configurations; in other cases, scale-out deployments achieve superior throughput. We also show that certain parallelism combinations, such as tensor with pipeline, lead to bandwidth underutilization due to inefficient data chunking, while increasing microbatch sizes beyond a certain point induces bursty execution and peak power excursions that worsen thermal throttling. These insights reveal how training performance is shaped by complex interactions between hardware, system topology, and model execution. We conclude by offering recommendations for system and hardware design to improve the scalability and reliability of future LLM systems and workloads. The source code of this project is available at https://github.com/sitar-lab/CharLLM-PPT.

This study investigates the register variation in texts written by humans and comparable texts produced by large language models (LLMs). Biber's multidimensional analysis (MDA) (Biber, 1988) is applied to a sample of human-written texts and AI-created texts generated to be their counterparts to find the dimensions of variation in which LLMs differ most significantly and most systematically from humans. As textual material, a new LLM-generated corpus AI-Brown is used, which is comparable to BE21 (a Brown family corpus representing contemporary British English). Since all languages except English are underrepresented in the training data of frontier LLMs, similar analysis is replicated on Czech using AI-Koditex corpus and Czech multidimensional model (Cvrˇ cek et al., 2018). Examined were 16 frontier models in various settings and prompts, with emphasis placed on the difference between base models and instruction-tuned models. Based on this, a benchmark is created through which models can be compared with each other and ranked in interpretable dimensions.

Traditional public opinion surveys face a number of challenges and risks related to measurement and representation dimensions, including, for example, coverage error due to incomplete frames and hard-to-reach groups, sampling error resulting from finite samples and complex designs, nonresponse error stemming from low participation and interview fatigue, measurement error introduced by questionnaire wording, and processing errors in coding and post-survey adjustments, among others (Groves, 1989; Groves and Lyberg, 2010; Weisberg, 2005). These errors could be amplified by substantial financial, human, and logistical demands, such as time spent on instrument design, piloting, and fieldwork that often forces a cost-quality trade-off that may distort population inferences. Consequently, there is a growing demand in the social sciences and market research for methods that reduce burden and cost while maintaining and improving overall data quality. Against this backdrop, Large Language Models (LLMs), trained extensively on vast and diverse data, emerge as promising alternatives for new research possibilities and applied research, including handling the abovementioned survey research limitations and measurement and representation errors. Indeed, recent advances in generative artificial intelligence (AI) suggest LLMs could serve for a number of classification tasks, including the creation of synthetic samples, providing simulated responses reflective of broader societal attitudes and behaviours (Argyle et al., 2023; Gilardi et al., 2023; González-Bustamante, 2024). The synthetic samples specifically may leverage the ability of LLMs 2 to generate contextually informed responses based on individual-level demographic characteristics and attitudes, and, in this way, potentially emulate public opinion without direct interaction with human respondents. This methodological innovation opens new avenues for rapid data collection, experimentation with sensitive topics, and a deeper understanding of complex public opinion dynamics that complement or even partially substitute for traditional surveys. Thus, the primary objective of this working paper is to evaluate the effectiveness and reliability of LLM-generated synthetic survey responses in reflecting real-world public opinion in Chile. Specifically, we aim to assess the predictive accuracy of a number of state-of-the-art private and open-source LLMs by comparing their synthetic respondents against human probabilistic responses.

The advancement of large language models (LLMs) and code agents has demonstrated significant potential to assist software engineering (SWE) tasks, such as autonomous issue resolution and feature addition. Existing AI for software engineering leaderboards (e.g., SWE-bench) focus solely on solution accuracy, ignoring the crucial factor of effectiveness in a resource-constrained world. This is a universal problem that also exists beyond software engineering tasks: any AI system should be more than correct - it must also be cost-effective. To address this gap, we introduce SWE-Effi, a set of new metrics to re-evaluate AI systems in terms of holistic effectiveness scores. We define effectiveness as the balance between the accuracy of outcome (e.g., issue resolve rate) and the resources consumed (e.g., token and time). In this paper, we specifically focus on the software engineering scenario by re-ranking popular AI systems for issue resolution on a subset of the SWE-bench benchmark using our new multi-dimensional metrics. We found that AI system's effectiveness depends not just on the scaffold itself, but on how well it integrates with the base model, which is key to achieving strong performance in a resource-efficient manner. We also identified systematic challenges such as the "token snowball" effect and, more significantly, a pattern of "expensive failures". In these cases, agents consume excessive resources while stuck on unsolvable tasks - an issue that not only limits practical deployment but also drives up the cost of failed rollouts during RL training. Lastly, we observed a clear trade-off between effectiveness under the token budget and effectiveness under the time budget, which plays a crucial role in managing project budgets and enabling scalable reinforcement learning, where fast responses are essential.

Timely and robust influenza incidence forecasting is critical for public health decision-making. This paper presents MAESTRO (Multi-modal Adaptive Estimation for Temporal Respiratory Disease Outbreak), a novel, unified framework that synergistically integrates advanced spectro-temporal modeling with multi-modal data fusion, including surveillance, web search trends, and meteorological data. By adaptively weighting heterogeneous data sources and decomposing complex time series patterns, the model achieves robust and accurate forecasts. Evaluated on over 11 years of Hong Kong influenza data (excluding the COVID-19 period), MAESTRO demonstrates state-of-the-art performance, achieving a superior model fit with an R-square of 0.956. Extensive ablations confirm the significant contributions of its multi-modal and spectro-temporal components. The modular and reproducible pipeline is made publicly available to facilitate deployment and extension to other regions and pathogens, presenting a powerful tool for epidemiological forecasting.

Recently, the physical capabilities of (M)LLMs have garnered increasing attention. However, existing benchmarks for physics suffer from two major gaps: they neither provide systematic and up-to-date coverage of real-world physics competitions such as physics Olympiads, nor enable direct performance comparison with humans. To bridge these gaps, we present HiPhO, the first benchmark dedicated to high school physics Olympiads with human-aligned evaluation. Specifically, HiPhO highlights three key innovations. (1) Comprehensive Data: It compiles 13 latest Olympiad exams from 2024-2025, spanning both international and regional competitions, and covering mixed modalities that encompass problems spanning text-only to diagram-based. (2) Professional Evaluation: We adopt official marking schemes to perform fine-grained grading at both the answer and step level, fully aligned with human examiners to ensure high-quality and domain-specific evaluation. (3) Comparison with Human Contestants: We assign gold, silver, and bronze medals to models based on official medal thresholds, thereby enabling direct comparison between (M)LLMs and human contestants. Our large-scale evaluation of 30 state-of-the-art (M)LLMs shows that: across 13 exams, open-source MLLMs mostly remain at or below the bronze level; open-source LLMs show promising progress with multiple golds; closed-source reasoning MLLMs can achieve 6 to 12 gold medals; and most models still have a significant gap from full marks. These results highlight the performance gap between open-source models and top students, the strong reasoning abilities of closed-source models, and the remaining room for improvement. HiPhO, a human-aligned Olympiad benchmark for multimodal physical reasoning, is open-source at https://github.com/SciYu/HiPhO with a public leaderboard at https://phyarena.github.io/.

Normalization layers are crucial for deep learning, but their Euclidean formulations are inadequate for data on manifolds. On the other hand, many Riemannian manifolds in machine learning admit gyro-structures, enabling principled extensions of Euclidean neural networks to non-Euclidean domains. Inspired by this, we introduce GyroBN, a principled Riemannian batch normalization framework for gyrogroups. We establish two necessary conditions, namely \emph{pseudo-reduction} and \emph{gyroisometric gyrations}, that guarantee GyroBN with theoretical control over sample statistics, and show that these conditions hold for all known gyrogroups in machine learning. Our framework also incorporates several existing Riemannian normalization methods as special cases. We further instantiate GyroBN on seven representative geometries, including the Grassmannian, five constant curvature spaces, and the correlation manifold, and derive novel gyro and Riemannian structures to enable these instantiations. Experiments across these geometries demonstrate the effectiveness of GyroBN. The code is available at https://github.com/GitZH-Chen/GyroBN.git.

The rise of model-sharing through frameworks and dedicated hubs makes Machine Learning significantly more accessible. Despite their benefits, these tools expose users to underexplored security risks, while security awareness remains limited among both practitioners and developers. To enable a more security-conscious culture in Machine Learning model sharing, in this paper we evaluate the security posture of frameworks and hubs, assess whether security-oriented mechanisms offer real protection, and survey how users perceive the security narratives surrounding model sharing. Our evaluation shows that most frameworks and hubs address security risks partially at best, often by shifting responsibility to the user. More concerningly, our analysis of frameworks advertising security-oriented settings and complete model sharing uncovered six 0-day vulnerabilities enabling arbitrary code execution. Through this analysis, we debunk the misconceptions that the model-sharing problem is largely solved and that its security can be guaranteed by the file format used for sharing. As expected, our survey shows that the surrounding security narrative leads users to consider security-oriented settings as trustworthy, despite the weaknesses shown in this work. From this, we derive takeaways and suggestions to strengthen the security of model-sharing ecosystems.

We introduce LongCat-Flash, a 560-billion-parameter Mixture-of-Experts (MoE) language model designed for both computational efficiency and advanced agentic capabilities. Stemming from the need for scalable efficiency, LongCat-Flash adopts two novel designs: (a) Zero-computation Experts, which enables dynamic computational budget allocation and activates 18.6B-31.3B (27B on average) per token depending on contextual demands, optimizing resource usage. (b) Shortcut-connected MoE, which enlarges the computation-communication overlap window, demonstrating notable gains in inference efficiency and throughput compared to models of a comparable scale. We develop a comprehensive scaling framework for large models that combines hyperparameter transfer, model-growth initialization, a multi-pronged stability suite, and deterministic computation to achieve stable and reproducible training. Notably, leveraging the synergy among scalable architectural design and infrastructure efforts, we complete model training on more than 20 trillion tokens within 30 days, while achieving over 100 tokens per second (TPS) for inference at a cost of \$0.70 per million output tokens. To cultivate LongCat-Flash towards agentic intelligence, we conduct a large-scale pre-training on optimized mixtures, followed by targeted mid- and post-training on reasoning, code, and instructions, with further augmentation from synthetic data and tool use tasks. Comprehensive evaluations demonstrate that, as a non-thinking foundation model, LongCat-Flash delivers highly competitive performance among other leading models, with exceptional strengths in agentic tasks. The model checkpoint of LongCat-Flash is open-sourced to foster community research. LongCat Chat: https://longcat.ai Hugging Face: https://huggingface.co/meituan-longcat GitHub: https://github.com/meituan-longcat