Topic model and document-clustering evaluations either use automated metrics that align poorly with human preferences or require expert labels that are intractable to scale. We design a scalable human evaluation protocol and a corresponding automated approximation that reflect practitioners' real-world usage of models. Annotators -- or an LLM-based proxy -- review text items assigned to a topic or cluster, infer a category for the group, then apply that category to other documents. Using this protocol, we collect extensive crowdworker annotations of outputs from a diverse set of topic models on two datasets. We then use these annotations to validate automated proxies, finding that the best LLM proxies are statistically indistinguishable from a human annotator and can therefore serve as a reasonable substitute in automated evaluations. Package, web interface, and data are at https://github.com/ahoho/proxann

In this paper, we propose an improved numerical algorithm for solving minimax problems based on nonsmooth optimization, quadratic programming and iterative process. We also provide a rigorous proof of convergence for our algorithm under some mild assumptions, such as gradient continuity and boundedness. Such an algorithm can be widely applied in various fields such as robust optimization, imbalanced learning, etc.

We present a novel framework for real-time sign language recognition using lightweight DNNs trained on limited data. Our system addresses key challenges in sign language recognition, including data scarcity, high computational costs, and discrepancies in frame rates between training and inference environments. By encoding sign language specific parameters, such as handshape, palm orientation, movement, and location into vectorized inputs, and leveraging MediaPipe for landmark extraction, we achieve highly separable input data representations. Our DNN architecture, optimized for sub 10MB deployment, enables accurate classification of 343 signs with less than 10ms latency on edge devices. The data annotation platform 'slait data' facilitates structured labeling and vector extraction. Our model achieved 92% accuracy in isolated sign recognition and has been integrated into the 'slait ai' web application, where it demonstrates stable inference.

Emerging AI technologies offer new opportunities to advance interventions that reduce polarization and promote political open-mindedness. We examined novel design strategies that leverage adaptive and emotionally-responsive civic narratives that may sustain students' emotional engagement in stories, and in turn, promote perspective-taking toward members of political out-groups. Drawing on theories from political psychology and narratology, we investigate how affective computing techniques can support three storytelling mechanisms: transportation into a story world, identification with characters, and interaction with the storyteller. Using a design-based research (DBR) approach, we iteratively developed and refined an AI-mediated Digital Civic Storytelling (AI-DCS) platform. Our prototype integrates facial emotion recognition and attention tracking to assess users' affective and attentional states in real time. Narrative content is organized around pre-structured story outlines, with beat-by-beat language adaptation implemented via GPT-4, personalizing linguistic tone to sustain students' emotional engagement in stories that center political perspectives different from their own. Our work offers a foundation for AI-supported, emotionally-sensitive strategies that address affective polarization while preserving learner autonomy. We conclude with implications for civic education interventions, algorithmic literacy, and HCI challenges associated with AI dialogue management and affect-adaptive learning environments.

Large language models (LLMs) are now widely accessible, reaching learners at all educational levels. This development has raised concerns that their use may circumvent essential learning processes and compromise the integrity of established assessment formats. In physics education, where problem solving plays a central role in instruction and assessment, it is therefore essential to understand the physics-specific problem-solving capabilities of LLMs. Such understanding is key to informing responsible and pedagogically sound approaches to integrating LLMs into instruction and assessment. This study therefore compares the problem-solving performance of a general-purpose LLM (GPT-4o, using varying prompting techniques) and a reasoning-optimized model (o1-preview) with that of participants of the German Physics Olympiad, based on a set of well-defined Olympiad problems. In addition to evaluating the correctness of the generated solutions, the study analyzes characteristic strengths and limitations of LLM-generated solutions. The findings of this study indicate that both tested LLMs (GPT-4o and o1-preview) demonstrate advanced problem-solving capabilities on Olympiad-type physics problems, on average outperforming the human participants. Prompting techniques had little effect on GPT-4o's performance, while o1-preview almost consistently outperformed both GPT-4o and the human benchmark. Based on these findings, the study discusses implications for the design of summative and formative assessment in physics education, including how to uphold assessment integrity and support students in critically engaging with LLMs.

The escalating computational costs of Large Language Model (LLM) inference have become a critical barrier to their widespread and sustainable deployment. While existing optimization strategies are effective, they are predominantly based on statistical heuristics or architectural modifications, lacking a guiding cognitive theory to manage the inference process itself. This paper aims to bridge this gap by introducing a novel paradigm: the Cognitive Load-Aware Inference (CLAI) framework, which operationalizes principles from Cognitive Load Theory (CLT) and neuroscience for LLM inference. We formalize the concepts of Intrinsic Cognitive Load, Extraneous Cognitive Load, and Germane Cognitive Load into quantifiable LLM metrics ($ICL_{LLM}$, $ECL_{LLM}$, and $GCL_{LLM}$), thereby reframing the inference process as a cognitive economics optimization problem: based on the intrinsic complexity of a problem ($ICL_{LLM}$), minimize wasteful computation ($ECL_{LLM}$), and strategically allocate the token budget to productive reasoning ($GCL_{LLM}$). We propose two implementation paths: CLAI-Prompt, a zero-shot method that guides a base LLM through cognitive control steps via a structured meta-prompt, and CLAI-Tune, a fine-tuned model that internalizes these principles for spontaneous cognitive economy. Across a range of benchmarks in complex reasoning, long-context question answering, and code generation, our methods achieve significant reductions in token consumption (up to 45\%) without sacrificing accuracy. Furthermore, CLAI-Tune exhibits an emergent ability to autonomously decompose difficult problems, a key characteristic of human expert cognition. This work demonstrates that by emulating the brain's resource management strategies, we can build more efficient, robust, and capable artificial intelligence systems.

Computer programming is undergoing a true transformation driven by powerful new tools for automatic source code generation based on large language models. This transformation is also manifesting in introductory programming courses at universities around the world, generating an in-depth debate about how programming content should be taught, learned, and assessed in the context of generative artificial intelligence. This article aims, on the one hand, to review the most relevant studies on this issue, highlighting the advantages and disadvantages identified in the specialized literature. On the other hand, it proposes enriching teaching and learning methodologies by focusing on code comprehension and execution rather than on mere coding or program functionality. In particular, it advocates for the use of visual representations of code and visual simulations of its execution as effective tools for teaching, learning, and assessing programming, thus fostering a deeper understanding among students. Finally, the opinions of students who took the object-oriented programming course are presented to provide preliminary context supporting the incorporation of visual simulations in Java (or other languages) as part of the training process.

Recent advances in reinforcement learning have shown that language models can develop sophisticated reasoning through training on tasks with verifiable rewards, but these approaches depend on human-curated problem-answer pairs and domain-specific reward engineering. We introduce SPIRAL, a self-play framework where models learn by playing multi-turn, zero-sum games against continuously improving versions of themselves, eliminating the need for human supervision. Through self-play, SPIRAL generates an infinite curriculum of progressively challenging problems as models must constantly adapt to stronger opponents. To enable this self-play training at scale, We implement a fully online, multi-turn, multi-agent reinforcement learning system for LLMs and propose role-conditioned advantage estimation (RAE) to stabilize multi-agent training. Using SPIRAL, self-play on zero-sum games produces reasoning capabilities that transfer broadly. Training Qwen3-4B-Base on Kuhn Poker alone achieves 8.6% improvement on math and 8.4% on general reasoning, outperforming SFT on 25,000 expert game trajectories. Analysis reveals that this transfer occurs through three cognitive patterns: systematic decomposition, expected value calculation, and case-by-case analysis. Multi-game training (TicTacToe, Kuhn Poker, Simple Negotiation) further enhances performance as each game develops distinct reasoning strengths. Applying SPIRAL to a strong reasoning model (DeepSeek-R1-Distill-Qwen-7B) can still lead to 2.0% average improvement. These results demonstrate that zero-sum games naturally develop transferable reasoning capabilities, highlighting a promising direction for autonomous reasoning development.

The influence of Artificial Intelligence (AI), and specifically Large Language Models (LLM), on education is continuously increasing. These models are frequently used by students, giving rise to the question whether current forms of assessment are still a valid way to evaluate student performance and comprehension. The theoretical framework developed in this paper is grounded in Constructive Alignment (CA) theory and Bloom's taxonomy for defining learning objectives. We argue that AI influences learning objectives of different Bloom levels in a different way, and assessment has to be adopted accordingly. Furthermore, in line with Bloom's vision, formative and summative assessment should be aligned on whether the use of AI is permitted or not. Although lecturers tend to agree that education and assessment need to be adapted to the presence of AI, a strong bias exists on the extent to which lecturers want to allow for AI in assessment. This bias is caused by a lecturer's familiarity with AI and specifically whether they use it themselves. To avoid this bias, we propose structured guidelines on a university or faculty level, to foster alignment among the staff. Besides that, we argue that teaching staff should be trained on the capabilities and limitations of AI tools. In this way, they are better able to adapt their assessment methods.

Water consumption is an increasingly critical dimension of computing sustainability, especially as AI workloads rapidly scale. However, current water impact assessment often overlooks where and when water stress is more severe. To fill in this gap, we present SCARF, the first general framework that evaluates water impact of computing by factoring in both spatial and temporal variations in water stress. SCARF calculates an Adjusted Water Impact (AWI) metric that considers both consumption volume and local water stress over time. Through three case studies on LLM serving, datacenters, and semiconductor fabrication plants, we show the hidden opportunities for reducing water impact by optimizing location and time choices, paving the way for water-sustainable computing. The code is available at https://github.com/jojacola/SCARF.