In-context learning (ICL)-the ability of transformer-based models to perform new tasks from examples provided at inference time-has emerged as a hallmark of modern language models. While recent works have investigated the mechanisms underlying ICL, its feasibility under formal privacy constraints remains largely unexplored. In this paper, we propose a differentially private pretraining algorithm for linear attention heads and present the first theoretical analysis of the privacy-accuracy trade-off for ICL in linear regression. Our results characterize the fundamental tension between optimization and privacy-induced noise, formally capturing behaviors observed in private training via iterative methods. Additionally, we show that our method is robust to adversarial perturbations of training prompts, unlike standard ridge regression. All theoretical findings are supported by extensive simulations across diverse settings.

Large Language Models (LLMs) have attracted significant attention in recommender systems for their excellent world knowledge capabilities. However, existing methods that rely on Euclidean space struggle to capture the rich hierarchical information inherent in textual and semantic data, which is essential for capturing user preferences. The geometric properties of hyperbolic space offer a promising solution to address this issue. Nevertheless, integrating LLMs-based methods with hyperbolic space to effectively extract and incorporate diverse hierarchical information is non-trivial. To this end, we propose a model-agnostic framework, named HyperLLM, which extracts and integrates hierarchical information from both structural and semantic perspectives. Structurally, HyperLLM uses LLMs to generate multi-level classification tags with hierarchical parent-child relationships for each item. Then, tag-item and user-item interactions are jointly learned and aligned through contrastive learning, thereby providing the model with clear hierarchical information. Semantically, HyperLLM introduces a novel meta-optimized strategy to extract hierarchical information from semantic embeddings and bridge the gap between the semantic and collaborative spaces for seamless integration. Extensive experiments show that HyperLLM significantly outperforms recommender systems based on hyperbolic space and LLMs, achieving performance improvements of over 40%. Furthermore, HyperLLM not only improves recommender performance but also enhances training stability, highlighting the critical role of hierarchical information in recommender systems.

Planning is central to agents and agentic AI. The ability to plan, e.g., creating travel itineraries within a budget, holds immense potential in both scientific and commercial contexts. Moreover, optimal plans tend to require fewer resources compared to ad-hoc methods. To date, a comprehensive understanding of existing planning benchmarks appears to be lacking. Without it, comparing planning algorithms' performance across domains or selecting suitable algorithms for new scenarios remains challenging. In this paper, we examine a range of planning benchmarks to identify commonly used testbeds for algorithm development and highlight potential gaps. These benchmarks are categorized into embodied environments, web navigation, scheduling, games and puzzles, and everyday task automation. Our study recommends the most appropriate benchmarks for various algorithms and offers insights to guide future benchmark development.

Large Multimodal Models (LMMs) uniformly perceive video frames, creating computational inefficiency for videos with inherently varying temporal information density. This paper present \textbf{Quicksviewer}, an LMM with new perceiving paradigm that partitions a video of nonuniform density into varying cubes using Gumbel Softmax, followed by a unified resampling for each cube to achieve efficient video understanding. This simple and intuitive approach dynamically compress video online based on its temporal density, significantly reducing spatiotemporal redundancy (overall 45$\times$ compression rate), while enabling efficient training with large receptive field. We train the model from a language backbone through three progressive stages, each incorporating lengthy videos on average of 420s/1fps thanks to the perceiving efficiency. With only 0.8M total video-text samples for training, our model outperforms the direct baseline employing a fixed partitioning strategy by a maximum of 8.72 in accuracy, demonstrating the effectiveness in performance. On Video-MME, Quicksviewer achieves SOTA under modest sequence lengths using just up to 5\% of tokens per frame required by baselines. With this paradigm, scaling up the number of input frames reveals a clear power law of the model capabilities. It is also empirically verified that the segments generated by the cubing network can help for analyzing continuous events in videos.

Construction workers face significant risks of work-related musculoskeletal disorders (WMSDs), driven by repetitive tasks, heavy load handling, and non-neutral postures in dynamic, unpredictable environments [1, 10]. In the U.S., construction workers experience an 11% higher WMSD rate than the average across industries, with the back and shoulders most affected [10]. While exoskeletons show promise in reducing physical strain--passive designs lowering back muscle activity by 10-40% and active ones achieving up to 80% reductions across multiple regions [5]--their practical deployment remains limited by discomfort and poor alignment with human movements, particularly in construction settings [6]. Central to these limitations is the challenge of accurately recognizing user intent across varied tasks, a gap that restricts effective collaboration [3, 34]. This misalignment heightens safety risks, as powered exoskeletons may generate destructive forces if their controlled output deviates from the user's intent [34]. Addressing this locomotion intent recognition challenge is pivotal to unlocking effective exoskeleton assistance in construction, particularly for diverse, safety-critical tasks like ladder climbing and obstacle navigation. Traditional evaluation of assistive technologies like lower-limb exoskeletons has focused narrowly on routine tasks such as straight walking [27], neglecting these critical locomotion modes and requiring a shift beyond conventional control paradigms that lack flexibility for dynamic contexts. Construction tasks are highly variable, requiring workers to adapt to shifting demands, irregular workflows, and unstructured environments where movement patterns are unpredictable [10]. This variability complicates the implementation of assistive technologies, as rigid control approaches struggle to accommodate rapid task transitions and environmental uncertainty.

Objective: Healthcare data fragmentation presents a major challenge for linking patient data, necessitating robust record linkage to integrate patient records from diverse sources. This study investigates the feasibility of leveraging language models for automated patient record linkage, focusing on two key tasks: blocking and matching. Materials and Methods: We utilized real-world healthcare data from the Missouri Cancer Registry and Research Center, linking patient records from two independent sources using probabilistic linkage as a baseline. A transformer-based model, RoBERTa, was fine-tuned for blocking using sentence embeddings. For matching, several language models were experimented under fine-tuned and zero-shot settings, assessing their performance against ground truth labels. Results: The fine-tuned blocking model achieved a 92% reduction in the number of candidate pairs while maintaining near-perfect recall. In the matching task, fine-tuned Mistral-7B achieved the best performance with only 6 incorrect predictions. Among zero-shot models, Mistral-Small-24B performed best, with a total of 55 incorrect predictions. Discussion: Fine-tuned language models achieved strong performance in patient record blocking and matching with minimal errors. However, they remain less accurate and efficient than a hybrid rule-based and probabilistic approach for blocking. Additionally, reasoning models like DeepSeek-R1 are impractical for large-scale record linkage due to high computational costs. Conclusion: This study highlights the potential of language models for automating patient record linkage, offering improved efficiency by eliminating the manual efforts required to perform patient record linkage. Overall, language models offer a scalable solution that can enhance data integration, reduce manual effort, and support disease surveillance and research.

AI assistants can impart value judgments that shape people's decisions and worldviews, yet little is known empirically about what values these systems rely on in practice. To address this, we develop a bottom-up, privacy-preserving method to extract the values (normative considerations stated or demonstrated in model responses) that Claude 3 and 3.5 models exhibit in hundreds of thousands of real-world interactions. We empirically discover and taxonomize 3,307 AI values and study how they vary by context. We find that Claude expresses many practical and epistemic values, and typically supports prosocial human values while resisting values like "moral nihilism". While some values appear consistently across contexts (e.g. "transparency"), many are more specialized and context-dependent, reflecting the diversity of human interlocutors and their varied contexts. For example, "harm prevention" emerges when Claude resists users, "historical accuracy" when responding to queries about controversial events, "healthy boundaries" when asked for relationship advice, and "human agency" in technology ethics discussions. By providing the first large-scale empirical mapping of AI values in deployment, our work creates a foundation for more grounded evaluation and design of values in AI systems.

Why do larger language models generalize better? To investigate this question, we develop generalization bounds on the pretraining objective of large language models (LLMs) in the compute-optimal regime, as described by the Chinchilla scaling laws. We introduce a novel, fully empirical Freedman-type martingale concentration inequality that tightens existing bounds by accounting for the variance of the loss function. This generalization bound can be decomposed into three interpretable components: the number of parameters per token, the loss variance, and the quantization error at a fixed bitrate. As compute-optimal language models are scaled up, the number of parameters per data point remains constant; however, both the loss variance and the quantization error decrease, implying that larger models should have smaller generalization gaps. We examine why larger models tend to be more quantizable from an information theoretic perspective, showing that the rate at which they can integrate new information grows more slowly than their capacity on the compute-optimal frontier. From these findings we produce a scaling law for the generalization gap, with bounds that become predictably stronger with scale.

Retrieval-augmented generation (RAG) enables large language models (LLMs) to generate answers with citations from source documents containing "ground truth", thereby reducing system hallucinations. A crucial factor in RAG evaluation is "support", whether the information in the cited documents supports the answer. To this end, we conducted a large-scale comparative study of 45 participant submissions on 36 topics to the TREC 2024 RAG Track, comparing an automatic LLM judge (GPT-4o) against human judges for support assessment. We considered two conditions: (1) fully manual assessments from scratch and (2) manual assessments with post-editing of LLM predictions. Our results indicate that for 56% of the manual from-scratch assessments, human and GPT-4o predictions match perfectly (on a three-level scale), increasing to 72% in the manual with post-editing condition. Furthermore, by carefully analyzing the disagreements in an unbiased study, we found that an independent human judge correlates better with GPT-4o than a human judge, suggesting that LLM judges can be a reliable alternative for support assessment. To conclude, we provide a qualitative analysis of human and GPT-4o errors to help guide future iterations of support assessment.

MILS (Multimodal Iterative LLM Solver) is a recently published framework that claims "LLMs can see and hear without any training" by leveraging an iterative, LLM-CLIP based approach for zero-shot image captioning. While this MILS approach demonstrates good performance, our investigation reveals that this success comes at a hidden, substantial computational cost due to its expensive multi-step refinement process. In contrast, alternative models such as BLIP-2 and GPT-4V achieve competitive results through a streamlined, single-pass approach. We hypothesize that the significant overhead inherent in MILS's iterative process may undermine its practical benefits, thereby challenging the narrative that zero-shot performance can be attained without incurring heavy resource demands. This work is the first to expose and quantify the trade-offs between output quality and computational cost in MILS, providing critical insights for the design of more efficient multimodal models.