Uncategorized

arXiv:2512.22955v1 Announce Type: new Abstract: Recent advancements have shown that reinforcement learning (RL) can substantially improve the reasoning abilities of large language models (LLMs). The effectiveness of such RL training, however, depends critically on the exploration space defined by the pre-trained model’s token-output distribution. In this paper, we revisit the standard cross-entropy loss, interpreting it as a specific instance of policy gradient optimization applied within a single-step episode. To systematically study how the pre-trained distribution shapes the exploration potential for subsequent RL, we propose a generalized pre-training objective that adapts on-policy RL principles to supervised learning. By framing next-token prediction as a stochastic decision process, we introduce a reward-shaping strategy that explicitly balances diversity and precision. Our method employs a positive reward scaling factor to control probability concentration on ground-truth tokens and a rank-aware mechanism that treats high-ranking and low-ranking negative tokens asymmetrically. This allows us to reshape the pre-trained token-output distribution and investigate how to provide a more favorable exploration space for RL, ultimately enhancing end-to-end reasoning performance. Contrary to the intuition that higher distribution entropy facilitates effective exploration, we find that imposing a precision-oriented prior yields a superior exploration space for RL.

arXiv:2512.22455v1 Announce Type: cross Abstract: Low-Rank Adaptation (LoRA) is a widely adopted parameter-efficient fine-tuning (PEFT) method. However, its linear adaptation process limits its expressive power. This means there is a gap between the expressive power of linear training and non-linear training. To bridge this gap, we propose AFA-LoRA, a novel training strategy that brings non-linear expressivity to LoRA while maintaining its seamless mergeability. Our key innovation is an annealed activation function that transitions from a non-linear to a linear transformation during training, allowing the adapter to initially adopt stronger representational capabilities before converging to a mergeable linear form. We implement our method on supervised fine-tuning, reinforcement learning, and speculative decoding. The results show that AFA-LoRA reduces the performance gap between LoRA and full-parameter training. This work enables a more powerful and practical paradigm of parameter-efficient adaptation.

arXiv:2512.22966v1 Announce Type: new Abstract: Large Language Models (LLMs) have demonstrated promise in medical knowledge assessments, yet their practical utility in real-world clinical decision-making remains underexplored. In this study, we evaluated the performance of three state-of-the-art LLMs-ChatGPT-4o, Gemini 1.5 Pro, and LIama 3.3 70B-in clinical decision support across the entire clinical reasoning workflow of a typical patient encounter. Using 36 case studies, we first assessed LLM’s out-of-the-box performance across five key sequential clinical decision-making tasks under two temperature settings (default vs. zero): differential diagnosis, essential immediate steps, relevant diagnostic testing, final diagnosis, and treatment recommendation. All models showed high variability by task, achieving near-perfect accuracy in final diagnosis, poor performance in relevant diagnostic testing, and moderate performance in remaining tasks. Furthermore, ChatGPT performed better under the zero temperature, whereas LIama showed stronger performance under the default temperature. Next, we assessed whether prompt engineering could enhance LLM performance by applying variations of the MedPrompt framework, incorporating targeted and random dynamic few-shot learning. The results demonstrate that prompt engineering is not a one-size-fit-all solution. While it significantly improved the performance on the task with lowest baseline accuracy (relevant diagnostic testing), it was counterproductive for others. Another key finding was that the targeted dynamic few-shot prompting did not consistently outperform random selection, indicating that the presumed benefits of closely matched examples may be counterbalanced by loss of broader contextual diversity. These findings suggest that the impact of prompt engineering is highly model and task-dependent, highlighting the need for tailored, context-aware strategies for integrating LLMs into healthcare.

arXiv:2508.07185v2 Announce Type: replace Abstract: Large Language Models (LLMs) suffer from a critical limitation: their knowledge is static and quickly becomes outdated. Retraining these massive models is computationally prohibitive, while existing knowledge editing techniques can be slow and may introduce unforeseen side effects. To address this, we propose DySK-Attn, a novel framework that enables LLMs to efficiently integrate real-time knowledge from a dynamic external source. Our approach synergizes an LLM with a dynamic Knowledge Graph (KG) that can be updated instantaneously. The core of our framework is a sparse knowledge attention mechanism, which allows the LLM to perform a coarse-to-fine grained search, efficiently identifying and focusing on a small, highly relevant subset of facts from the vast KG. This mechanism avoids the high computational cost of dense attention over the entire knowledge base and mitigates noise from irrelevant information. We demonstrate through extensive experiments on time-sensitive question-answering tasks that DySK-Attn significantly outperforms strong baselines, including standard Retrieval-Augmented Generation (RAG) and model editing techniques, in both factual accuracy for updated knowledge and computational efficiency. Our framework offers a scalable and effective solution for building LLMs that can stay current with the ever-changing world.

arXiv:2508.10426v2 Announce Type: replace Abstract: Large language models (LLMs) are limited by substantial computational cost. We introduce a “computational economics” framework that treats an LLM as an internal economy of resource-constrained agents (attention heads and neuron blocks) that must allocate scarce computation to maximize task utility. First, we show empirically that when computation is scarce, standard LLMs reallocate attention toward high-value tokens while preserving accuracy. Building on this observation, we propose an incentive-driven training paradigm that augments the task loss with a differentiable computation cost term, encouraging sparse and efficient activations. On GLUE (MNLI, STS-B, CoLA) and WikiText-103, the method yields a family of models that trace a Pareto frontier and consistently dominate post-hoc pruning; for a similar accuracy we obtain roughly a forty percent reduction in FLOPS and lower latency, together with more interpretable attention patterns. These results indicate that economic principles offer a principled route to designing efficient, adaptive, and more transparent LLMs under strict resource constraints.

arXiv:2509.10436v2 Announce Type: replace Abstract: To optimize the reasoning and problem-solving capabilities of Large Language Models (LLMs), we propose a novel cloud-edge collaborative architecture that enables a structured multi-agent prompting framework. This framework comprises three specialized components: GuideLLM, a lightweight model deployed at the edge to provide methodological guidance; SolverLLM, a more powerful model hosted in the cloud and responsible for generating code solutions; and JudgeLLM, an automated evaluator for assessing solution correctness and quality. To evaluate and demonstrate the effectiveness of this architecture in realistic settings, we introduce RefactorCoderQA, a comprehensive benchmark designed to evaluate and enhance the performance of LLMs across multi-domain coding tasks. Motivated by the limitations of existing benchmarks, RefactorCoderQA systematically covers multiple technical domains, including Software Engineering, Data Science, Machine Learning, and Natural Language Processing, using authentic coding challenges sourced from Stack Overflow. We propose RefactorCoder-MoE, a fine-tuned mixture-of-experts (MoE) code language model based on DeepSeek-Coder-7B-Instruct, adapted to the RefactorCoderQA benchmark using QLoRA for domain-specific coding question answering. Extensive experiments demonstrate that RefactorCoder-MoE achieves strong and competitive performance, significantly outperforming all evaluated open-source and commercial baselines, with an overall accuracy of 76.84%.

arXiv:2411.02455v2 Announce Type: replace Abstract: Course evaluation plays a critical role in ensuring instructional quality and guiding curriculum development in higher education. However, traditional evaluation methods, such as student surveys, classroom observations, and expert reviews, are often constrained by subjectivity, high labor costs, and limited scalability. With recent advancements in large language models (LLMs), new opportunities have emerged for generating consistent, fine-grained, and scalable course evaluations. This study investigates the use of three representative LLMs for automated course evaluation at both the micro level (classroom discussion analysis) and the macro level (holistic course review). Using classroom interaction transcripts and a dataset of 100 courses from a major institution in China, we demonstrate that LLMs can extract key pedagogical features and generate structured evaluation results aligned with expert judgement. A fine-tuned version of Llama shows superior reliability, producing score distributions with greater differentiation and stronger correlation with human evaluators than its counterparts. The results highlight three major findings: (1) LLMs can reliably perform systematic and interpretable course evaluations at both the micro and macro levels; (2) fine-tuning and prompt engineering significantly enhance evaluation accuracy and consistency; and (3) LLM-generated feedback provides actionable insights for teaching improvement. These findings illustrate the promise of LLM-based evaluation as a practical tool for supporting quality assurance and educational decision-making in large-scale higher education settings.

arXiv:2512.21789v1 Announce Type: new Abstract: Between 2021 and 2025, the SciCap project grew from a small seed-funded idea at The Pennsylvania State University (Penn State) into one of the central efforts shaping the scientific figure-captioning landscape. Supported by a Penn State seed grant, Adobe, and the Alfred P. Sloan Foundation, what began as our attempt to test whether domain-specific training, which was successful in text models like SciBERT, could also work for figure captions expanded into a multi-institution collaboration. Over these five years, we curated, released, and continually updated a large collection of figure-caption pairs from arXiv papers, conducted extensive automatic and human evaluations on both generated and author-written captions, navigated the rapid rise of large language models (LLMs), launched annual challenges, and built interactive systems that help scientists write better captions. In this piece, we look back at the first five years of SciCap and summarize the key technical and methodological lessons we learned. We then outline five major unsolved challenges and propose directions for the next phase of research in scientific figure captioning.

It’s been a busy and productive year here at MIT Technology Review. We published magazine issues on power, creativity, innovation, bodies, relationships, and security. We hosted 14 exclusive virtual conversations with our editors and outside experts in our subscriber-only series, Roundtables, and held two events on MIT’s campus. And we published hundreds of articles online, following new developments in computing, climate tech, robotics, and more.  As the year winds down, we wanted to give you a chance to revisit a bit of this work with us. Whether we were covering the red-hot rise of artificial intelligence or the future of biotech, these are some of the stories that resonated the most with our readers.  We did the math on AI’s energy footprint. Here’s the story you haven’t heard. Understanding AI’s energy use was a huge global conversation in 2025 as hundreds of millions of people began using generative AI tools on a regular basis. Senior reporters James O’Donnell and Casey Crownhart dug into the numbers and published an unprecedented look at AI’s resource demand, down to the level of a single query, to help us know how much energy and water AI may require moving forward.  We’re learning more about what vitamin D does to our bodies Vitamin D deficiency is widespread, particularly in the winter when there’s less sunlight to drive its production in our bodies. The “sunshine vitamin” is important for bone health, but as senior reporter Jessica Hamzelou reported, recent research is also uncovering surprising new insights into other ways it might influence our bodies, including our immune systems and heart health. What is AI? Senior editor Will Douglas Heaven’s expansive look at how to define AI was published in 2024, but it still managed to connect with many readers this year. He lays out why no one can agree on what AI is—and explains why that ambiguity matters, and how it can inform our own critical thinking about this technology. Ethically sourced “spare” human bodies could revolutionize medicine In this thought-provoking op-ed, a team of experts at Stanford University argue that creating living human bodies that can’t think, don’t have any awareness, and can’t feel pain could shake up medical research and drug development by providing essential biological materials for testing and transplantation. Recent advances in biotechnology now provide a potential pathway to such “bodyoids,” though plenty of technical challenges and ethical hurdles remain.  It’s surprisingly easy to stumble into a relationship with an AI chatbot Chatbots were everywhere this year, and reporter Rhiannon Williams chronicled how quickly people can develop bonds with one. That’s all right for some people, she notes, but dangerous for others. Some folks even describe unintentionally forming romantic relationships with chatbots. This is a trend we’ll definitely be keeping an eye on in 2026.  Is this the electric grid of the future? The electric grid is bracing for disruption from more frequent storms and fires, as well as an uncertain policy and regulatory landscape. And in many ways, the publicly owned utility company Lincoln Electric in Nebraska is an ideal lens through which to examine this shift as it works through the challenges of delivering service that’s reliable, affordable, and sustainable. Exclusive: A record-breaking baby has been born from an embryo that’s over 30 years old This year saw the birth of the world’s “oldest baby”: Thaddeus Daniel Pierce, who arrived on July 26. The embryo he developed from was created in 1994 during the early days of IVF and had been frozen and sitting in storage ever since. The new baby’s parents were toddlers at the time, and the embryo was donated to them decades later via a Christian “embryo adoption” agency.   How these two brothers became go-to experts on America’s “mystery drone” invasion Twin brothers John and Gerald Tedesco teamed up to investigate a concerning new threat—unidentified drones. In 2024 alone, some 350 drones entered airspace over a hundred different US military installations, and many cases went unsolved, according to a top military official. This story takes readers inside the equipment-filled RV the Tedescos created to study mysterious aerial phenomena, and how they made a name for themselves among government officials.  10 Breakthrough Technologies of 2025  Our newsroom has published this annual look at advances that will matter in the long run for over 20 years. This year’s list featured generative AI search, cleaner jet fuel, long-acting HIV prevention meds, and other emerging technologies that our journalists think are worth watching. We’ll publish the 2026 edition of the list on January 12, so stay tuned. (In the meantime, here’s what didn’t make the cut.)