Recent advances in autonomous digital agents from industry (e.g., Manus AI and Gemini's research mode) highlight potential for structured tasks by autonomous decision-making and task decomposition; however, it remains unclear to what extent the agent-based systems can improve multimodal web data productivity. We study this in the realm of journalism, which requires iterative planning, interpretation, and contextual reasoning from multimodal raw contents to form a well structured news. We introduce NEWSAGENT, a benchmark for evaluating how agents can automatically search available raw contents, select desired information, and edit and rephrase to form a news article by accessing core journalistic functions. Given a writing instruction and firsthand data as how a journalist initiates a news draft, agents are tasked to identify narrative perspectives, issue keyword-based queries, retrieve historical background, and generate complete articles. Unlike typical summarization or retrieval tasks, essential context is not directly available and must be actively discovered, reflecting the information gaps faced in real-world news writing. NEWSAGENT includes 6k human-verified examples derived from real news, with multimodal contents converted to text for broad model compatibility. We evaluate open- and closed-sourced LLMs with commonly-used agentic frameworks on NEWSAGENT, which shows that agents are capable of retrieving relevant facts but struggling with planning and narrative integration. We believe that NEWSAGENT serves a realistic testbed for iterating and evaluating agent capabilities in terms of multimodal web data manipulation to real-world productivity.

Large Language Models (LLMs) achieve gold-medal performance across many benchmarks, yet it remains unclear whether such success reflects genuine reasoning or pattern matching. From a cognitive science perspective, an informative test is whether models can master an unfamiliar language through explicit metalinguistic deductive learning, a paradigm where human learners can reliably internalise grammatical systems through metalinguistic reasoning. We address this question with Camlang, a novel constructed language that exhibits naturalistic yet unattested feature combinations. Camlang consists of two explicit resources, a grammar book and a bilingual dictionary, which mirror adult second-language learning via explicit grammar rules and lexical lookup, and enable us to disentangle errors in morpho-syntax, lexical semantics, and sentence-level reasoning. Human experiments show that these resources are sufficient for participants to acquire Camlang and successfully solve Camlang tasks. To operationalise evaluation, we adapt CommonsenseQA into Camlang, creating Camlang-CSQA-v0, the first task in a broader suite where solving questions requires applying grammar rules and lexical mappings. Experimental results show that GPT-5 achieves 98\% EM accuracy in English but only 47\% in Camlang, far below human performance at 87\%, while other state-of-the-art reasoning LLMs perform even worse. Human verification further reveals that most model successes stem from shallow lexical alignment while GPT-5 shows emerging metalinguistic awareness to a limited extent but not systematic grammatical mastery as humans. Camlang establishes a cognitively grounded evaluation paradigm that exposes fundamental gaps between current models and human metalinguistic competence.

Despite the empirical success of prompt tuning in adapting pretrained language models to new tasks, theoretical analyses of its capabilities remain limited. Existing theoretical work primarily addresses universal approximation properties, demonstrating results comparable to standard weight tuning. In this paper, we explore a different aspect of the theory of transformers: the memorization capability of prompt tuning. We provide two principal theoretical contributions. First, we prove that the amount of information memorized by a transformer cannot scale faster than linearly with the prompt length. Second, and more importantly, we present the first formal proof of a phenomenon empirically observed in large language models: performance degradation in transformers with extended contexts. We rigorously demonstrate that transformers inherently have limited memory, constraining the amount of information they can retain, regardless of the context size. This finding offers a fundamental understanding of the intrinsic limitations of transformer architectures, particularly their ability to handle long sequences.

In the digital age, people often turn to the Internet in search of medical advice and recommendations. With the increasing volume of online content, it has become difficult to distinguish reliable sources from misleading information. Similarly, millions of medical studies are published every year, making it challenging for researchers to keep track of the latest scientific findings. These evolving studies can reach differing conclusions, which is not reflected in traditional search tools. To address these challenges, we introduce MedSEBA, an interactive AI-powered system for synthesizing evidence-based answers to medical questions. It utilizes the power of Large Language Models to generate coherent and expressive answers, but grounds them in trustworthy medical studies dynamically retrieved from the research database PubMed. The answers consist of key points and arguments, which can be traced back to respective studies. Notably, the platform also provides an overview of the extent to which the most relevant studies support or refute the given medical claim, and a visualization of how the research consensus evolved through time. Our user study revealed that medical experts and lay users find the system usable and helpful, and the provided answers trustworthy and informative. This makes the system well-suited for both everyday health questions and advanced research insights.

Gradient-based adversarial prompting, such as the Greedy Coordinate Gradient (GCG) algorithm, has emerged as a powerful method for jailbreaking large language models (LLMs). In this paper, we present a systematic appraisal of GCG and its annealing-augmented variant, T-GCG, across open-source LLMs of varying scales. Using Qwen2.5-0.5B, LLaMA-3.2-1B, and GPT-OSS-20B, we evaluate attack effectiveness on both safety-oriented prompts (AdvBench) and reasoning-intensive coding prompts. Our study reveals three key findings: (1) attack success rates (ASR) decrease with model size, reflecting the increasing complexity and non-convexity of larger models' loss landscapes; (2) prefix-based heuristics substantially overestimate attack effectiveness compared to GPT-4o semantic judgments, which provide a stricter and more realistic evaluation; and (3) coding-related prompts are significantly more vulnerable than adversarial safety prompts, suggesting that reasoning itself can be exploited as an attack vector. In addition, preliminary results with T-GCG show that simulated annealing can diversify adversarial search and achieve competitive ASR under prefix evaluation, though its benefits under semantic judgment remain limited. Together, these findings highlight the scalability limits of GCG, expose overlooked vulnerabilities in reasoning tasks, and motivate further development of annealing-inspired strategies for more robust adversarial evaluation.

Efficient Key-Value (KV) cache management is essential for processing long text sequences in large language models (LLMs), where memory constraints often limit performance. Conventional KV eviction strategies, such as top-k selection based on attention scores, depend on static heuristics that fail to capture the evolving implicit dependencies among tokens during inference. To overcome this, we propose GraphKV, a graph-based framework that redefines token selection for KV cache compression. In GraphKV, tokens are modeled as nodes with importance scores, and edges represent their similarity relationships. Through a decay-signal-propagation mechanism, token importance is dynamically updated by propagating information across the graph, enabling adaptive retention of the most contextually significant tokens. GraphKV can be seamlessly utilized in existing KV cache eviction methods such as SnapKV and PyramidKV in a plug-and-play manner. Codes will be released on Github.

Large language models (LLMs) are increasingly expected to go beyond simple factual queries toward Deep Research--tasks that require decomposing questions into sub-problems, coordinating multi-step reasoning, and synthesizing evidence from diverse sources. We formalize Deep Research tasks with verifiable answers as Hierarchical Constraint Satisfaction Problems (HCSPs), which are fundamentally different from single-constraint, multi-hop, or flat CSP formulations. However, existing benchmarks (e.g., Natural Questions, HotpotQA) fail to capture this complexity, while recent synthetic datasets often introduce shortcut reasoning, knowledge leakage, or lack sufficient structural depth. To address this gap, we introduce InfoSeek, a scalable framework for synthesizing complex Deep Research tasks. InfoSeek uses a dual-agent system to recursively build a Research Tree from large-scale webpages, blurring intermediate nodes into valid sub-problems, and converting these trees into natural language questions that require traversing the full hierarchy. It also enables rapid scaling, yielding over 50K training examples, a curated test set, and reasoning trajectories generated via reject sampling. Experiments show that models trained on InfoSeek consistently outperform strong baselines. On a challenging benchmark BrowseComp-Plus, 3B LLMs optimized with InfoSeek surpass much larger 32B models and lightweight commercial APIs (e.g., Gemini2.5-Flash), while achieving performance comparable to stronger APIs (e.g., Gemini2.5-Pro). By preserving meta-information such as intermediate steps and retrieval labels, InfoSeek further supports advanced optimization strategies, including compound reward design and trajectory-level exploration. We provide our codes and datasets in this repository.Figure 1: Performance comparison on the BrowseComp-Plus benchmark. InfoSeeker-3B, a compact LLM trained with the InfoSeek dataset, significantly outperforms Qwen3-32B and achieves performance on par with leading commercial LLMs (ordered in API prices), highlighting the strong potential of InfoSeek for advancing Deep Research tasks.

Despite recent progress, Graphic User Interface (GUI) agents powered by Large Language Models (LLMs) struggle with complex mobile tasks due to limited app-specific knowledge. While UI Transition Graphs (UTGs) offer structured navigation representations, they are underutilized due to poor extraction and inefficient integration. We introduce KG-RAG, a Knowledge Graph-driven Retrieval-Augmented Generation framework that transforms fragmented UTGs into structured vector databases for efficient real-time retrieval. By leveraging an intent-guided LLM search method, KG-RAG generates actionable navigation paths, enhancing agent decision-making. Experiments across diverse mobile apps show that KG-RAG outperforms existing methods, achieving a 75.8% success rate (8.9% improvement over AutoDroid), 84.6% decision accuracy (8.1% improvement), and reducing average task steps from 4.5 to 4.1. Additionally, we present KG-Android-Bench and KG-Harmony-Bench, two benchmarks tailored to the Chinese mobile ecosystem for future research. Finally, KG-RAG transfers to web/desktop (+40% SR on Weibo-web; +20% on QQ Music-desktop), and a UTG cost ablation shows accuracy saturates at ~4h per complex app, enabling practical deployment trade-offs.

Surgical video understanding is crucial for facilitating Computer-Assisted Surgery (CAS) systems. Despite significant progress in existing studies, two major limitations persist, including inadequate visual content perception and insufficient temporal awareness in surgical videos, and hinder the development of versatile CAS solutions. In this work, we propose the SurgLLM framework, an effective large multimodal model tailored for versatile surgical video understanding tasks with enhanced spatial focus and temporal awareness. Specifically, to empower the spatial focus of surgical videos, we first devise Surgical Context-aware Multimodal Pretraining (Surg-Pretrain) for the video encoder of SurgLLM, by performing instrument-centric Masked Video Reconstruction (MV-Recon) and subsequent multimodal alignment. To incorporate surgical temporal knowledge into SurgLLM, we further propose Temporal-aware Multimodal Tuning (TM-Tuning) to enhance temporal reasoning with interleaved multimodal embeddings. Moreover, to accommodate various understanding tasks of surgical videos without conflicts, we devise a Surgical Task Dynamic Ensemble to efficiently triage a query with optimal learnable parameters in our SurgLLM. Extensive experiments performed on diverse surgical video understanding tasks, including captioning, general VQA, and temporal VQA, demonstrate significant improvements over the state-of-the-art approaches, validating the effectiveness of our SurgLLM in versatile surgical video understanding. The source code is available at https://github.com/franciszchen/SurgLLM.

Reinforcement Learning (RL) is known for its strong decision-making capabilities and has been widely applied in various real-world scenarios. However, with the increasing availability of offline datasets and the lack of well-designed online environments from human experts, the challenge of generalization in offline RL has become more prominent. Due to the limitations of offline data, RL agents trained solely on collected experiences often struggle to generalize to new tasks or environments. To address this challenge, we propose LLM-Driven Policy Diffusion (LLMDPD), a novel approach that enhances generalization in offline RL using task-specific prompts. Our method incorporates both text-based task descriptions and trajectory prompts to guide policy learning. We leverage a large language model (LLM) to process text-based prompts, utilizing its natural language understanding and extensive knowledge base to provide rich task-relevant context. Simultaneously, we encode trajectory prompts using a transformer model, capturing structured behavioral patterns within the underlying transition dynamics. These prompts serve as conditional inputs to a context-aware policy-level diffusion model, enabling the RL agent to generalize effectively to unseen tasks. Our experimental results demonstrate that LLMDPD outperforms state-of-the-art offline RL methods on unseen tasks, highlighting its effectiveness in improving generalization and adaptability in diverse settings.