Tokenization is the first -- and often least scrutinized -- step of most NLP pipelines. Standard algorithms for learning tokenizers rely on frequency-based objectives, which favor languages dominant in the training data and consequently leave lower-resource languages with tokenizations that are disproportionately longer, morphologically implausible, or even riddled with placeholders. This phenomenon ultimately amplifies computational and financial inequalities between users from different language backgrounds. To remedy this, we introduce Parity-aware Byte Pair Encoding (BPE), a variant of the widely-used BPE algorithm. At every merge step, Parity-aware BPE maximizes the compression gain of the currently worst-compressed language, trading a small amount of global compression for cross-lingual parity. We find empirically that Parity-aware BPE leads to more equitable token counts across languages, with negligible impact on global compression rate and no substantial effect on language-model performance in downstream tasks.

Multi-turn problem solving is critical yet challenging for Large Reasoning Models (LRMs) to reflect on their reasoning and revise from feedback. Existing Reinforcement Learning (RL) methods train large reasoning models on a single-turn paradigm with verifiable rewards. However, we observe that models trained with existing RL paradigms often lose their ability to solve problems across multiple turns and struggle to revise answers based on contextual feedback, leading to repetitive responses. We ask: can LRMs learn to reflect their answers in a multi-turn context? In this work, we find that training models with multi-turn RL using only unary feedback (e.g., "Let's try again") after wrong answers can improve both single-turn performance and multi-turn reasoning. We introduce Unary Feedback as Observation (UFO) for reinforcement learning, which uses minimal yet common unary user feedback during iterative problem solving. It can be easily applied to existing single-turn RL training setups. Experimental results show that RL training with UFO keeps single-turn performance and improves multi-turn reasoning accuracy by up to 14%, enabling language models to better react to feedback in multi-turn problem solving. To further minimize the number of turns needed for a correct answer while encouraging diverse reasoning when mistakes occur, we design reward structures that guide models to produce careful and deliberate answers in each turn. Code: https://github.com/lichengliu03/unary-feedback

Geometry problem solving, a crucial aspect of mathematical reasoning, is vital across various domains, including education, the assessment of AI's mathematical abilities, and multimodal capability evaluation. The recent surge in deep learning technologies, particularly the emergence of multimodal large language models, has significantly accelerated research in this area. This paper provides a survey of the applications of deep learning in geometry problem solving, including (i) a comprehensive summary of the relevant tasks in geometry problem solving; (ii) a thorough review of related deep learning methods; (iii) a detailed analysis of evaluation metrics and methods; and (iv) a critical discussion of the current challenges and future directions that can be explored. Our objective is to offer a comprehensive and practical reference of deep learning for geometry problem solving, thereby fostering further advancements in this field. We create a continuously updated list of papers on GitHub: https://github.com/majianz/dl4gps.

While Large Reasoning Models (LRMs) have demonstrated success in complex reasoning tasks through long chain-of-thought (CoT) reasoning, their inference often involves excessively verbose reasoning traces, resulting in substantial inefficiency. To address this, we propose Distilled Reasoning Pruning (DRP), a hybrid framework that combines inference-time pruning with tuning-based distillation, two widely used strategies for efficient reasoning. DRP uses a teacher model to perform skill-aware step decomposition and content pruning, and then distills the pruned reasoning paths into a student model, enabling it to reason both efficiently and accurately. Across several challenging mathematical reasoning datasets, we find that models trained with DRP achieve substantial improvements in token efficiency without sacrificing accuracy. Specifically, DRP reduces average token usage on GSM8K from 917 to 328 while improving accuracy from 91.7% to 94.1%, and achieves a 43% token reduction on AIME with no performance drop. Further analysis shows that aligning the reasoning structure of training CoTs with the student's reasoning capacity is critical for effective knowledge transfer and performance gains.

Large Language Models (LLMs) have demonstrated significant promise for various applications in healthcare. However, their effectiveness in the Arabic medical domain remains unexplored due to the lack of high-quality domain-specific datasets and benchmarks. This study introduces MedArabiQ, a new benchmark dataset consisting of seven Arabic medical tasks, covering multiple specialties and including multiple-choice questions, fill-in-the-blank questions, and patient-doctor questions and answers. We first constructed the dataset using past medical exams as well as publicly available datasets. We conducted an extensive evaluation with eight state-of-the-art open-access and proprietary high-resource LLMs, including GPT-4, Deepseek v3, and Gemini 1.5. Our findings highlight the need for the creation of new high-quality benchmarks that span different languages to ensure fair deployment and scalability of LLMs in healthcare. By establishing this benchmark and releasing the dataset, we provide a foundation for future research aimed at evaluating and enhancing the multilingual capabilities of LLMs for the equitable use of generative AI in healthcare. Data Availability In this article, we present a new benchmark dataset, MedArabiQ, designed to evaluate the performance of LLMs on Arabic medical tasks.

Large language models (LLMs) are commonly trained on multi-domain datasets, where domain sampling strategies significantly impact model performance due to varying domain importance across downstream tasks. Existing approaches for optimizing domain-level sampling strategies struggle with maintaining intra-domain consistency and accurately measuring domain impact. In this paper, we present Domain Impact-aware Data Sampling (DIDS). To ensure intra-domain consistency, a gradient clustering algorithm is proposed to group training data based on their learning effects, where a proxy language model and dimensionality reduction are employed to reduce computational overhead. To accurately measure domain impact, we develop a Fisher Information Matrix (FIM) guided metric that quantifies how domain-specific parameter updates affect the model's output distributions on downstream tasks, with theoretical guarantees. Furthermore, to determine optimal sampling ratios, DIDS combines both the FIM-guided domain impact assessment and loss learning trajectories that indicate domain-specific potential, while accounting for diminishing marginal returns. Extensive experiments demonstrate that DIDS achieves 3.4% higher average performance while maintaining comparable training efficiency. The code is available at https://github.com/shiweijiezero/DIDS.

I'm embarrassed to admit this in my mid-forties, but I've never understood chess well enough to play a full game. My son and daughter both learned how to play in elementary school. I was glad they had that experience. I tried to pick up the game when they did, but, as a busy mom of three little kids, I just didn't have the time, the interest, or the stamina to really sit down and learn. Chess became more popular during the pandemic, and the boom has stuck around; according to a recent Yougov.com

Membership in ACM includes a subscription to Communications of the ACM (CACM), the computing industry's most trusted source for staying connected to the world of advanced computing. Once artificial intelligence is added to augmented and virtual reality devices, it could take the technology mainstream. Meta CEO Mark Zuckerberg has spent billions of dollars year after year to develop augmented reality (AR) and virtual reality (VR) technologies, without much financial return to show for it. The company has spent north of $80 billion since 2014 on the technologies since acquiring Oculus, a VR hardware startup, including $20 billion in 2024 alone. In that time, it has lost billions on AR/VR, with products like its Quest headset consistently operating at a loss.

Everyone has a stake in how tech is shaping education today. From the tech moguls and venture capitalists who are starting "microschools" and building ed-tech tools to policymakers who are writing bills to safeguard kids online and teachers who are getting creative about using AI for school. WIRED explored all this and more in our recent back-to-school digital edition, and we're excited to talk about it at our next subscriber-only livestream on Thursday, August 28, at 1 pm ET / 10 am PT / 6 pm UK. We'll talk about what we learned--turns out that AI has given new life to a skill many people thought would be extinct soon, and one country is attempting to actually ban social media for teenagers--but we also want to hear from you. Share your stories and questions for us via this form, or leave them in the comments below.

Yesterday, we covered Google's report that a typical query to its Gemini AI consumes only "five drops of water." That figure is now facing criticism from several AI experts, according to The Verge… and that includes one of the authors of one of the reports referred to by Google. AI researcher Shaolei Ren--a professor at University of California Riverside and one of the authors of the report Making AI Less "Thirsty": Uncovering and Addressing the Secret Water Footprint of AI Models--previously estimated that Microsoft's data center consumed 700,000 liters of water to train OpenAI's GPT-3 model. He also calculated that a ChatGPT conversation of 20 to 50 messages can consume close to a pint of water, which is far more than Google's estimate. Ren and other AI researchers argue that Google is wrong to leave out the indirect water consumption of its AI models.