--Music editing has emerged as an important and practical area of artificial intelligence, with applications ranging from video game and film music production to personalizing existing tracks according to user preferences. However, existing models face significant limitations, such as being restricted to editing synthesized music generated by their own models, requiring highly precise prompts, or necessitating task-specific retraining--thus lacking true zero-shot capability. Experimental results demonstrate that our approach outperforms existing methods in preserving musical content, structural consistency, and editing fidelity, establishing a strong foundation for controllable music editing in real-world scenarios. The landscape of audio generation has shifted dramatically in recent years. Text-to-music systems now allow users to compose entire musical pieces from simple textual descriptions, powered by advances in diffusion models and transformer architectures [1]-[11]. While impressive, these systems are still primarily designed for creation from scratch . In contrast, real-world music practice often revolves around editing: refining a performance, altering instrumentation, or adapting an existing recording into a new style. For musicians, producers, and casual creators alike, the ability to reshape existing audio is often more valuable than generating entirely new material. Music editing, however, is fundamentally more difficult than generation. It requires the model to balance two competing goals: applying the requested modification faithfully, and preserving the rich details of the input recording that should remain unchanged. This trade-off is especially challenging when dealing with expressive, polyphonic, or multi-instrumental recordings. Existing research has attempted to address editing through supervised datasets of paired "before" and "after" examples [12]-[14], or through zero-shot latent manipulations in diffusion models [15]-[17]. Y et most methods remain restricted by their limitation to specific editing tasks, operate mainly on model-generated music rather than arbitrary recordings, and often require very precise prompts to succeed [15], [17].

The emergence of Large Language Models (LLMs) demonstrates their potential to encapsulate the logic and patterns inherent in human behavior simulation by leveraging extensive web data pre-training. However, the boundaries of LLM capabilities in social simulation remain unclear. To further explore the social attributes of LLMs, we introduce the CiteAgent framework, designed to generate citation networks based on human-behavior simulation with LLM-based agents. CiteAgent successfully captures predominant phenomena in real-world citation networks, including power-law distribution, citational distortion, and shrinking diameter. Building on this realistic simulation, we establish two LLM-based research paradigms in social science: LLM-SE (LLM-based Survey Experiment) and LLM-LE (LLM-based Laboratory Experiment). These paradigms facilitate rigorous analyses of citation network phenomena, allowing us to validate and challenge existing theories. Additionally, we extend the research scope of traditional science of science studies through idealized social experiments, with the simulation experiment results providing valuable insights for real-world academic environments. Our work demonstrates the potential of LLMs for advancing science of science research in social science.

Large language models (LLMs) are increasingly integrated into IT infrastructures, where they process user data according to predefined instructions. However, conventional LLMs remain vulnerable to prompt injection, where malicious users inject directive tokens into the data to subvert model behavior. Existing defenses train LLMs to semantically separate data and instruction tokens, but still struggle to (1) balance utility and security and (2) prevent instruction-like semantics in the data from overriding the intended instructions. We propose DRIP, which (1) precisely removes instruction semantics from tokens in the data section while preserving their data semantics, and (2) robustly preserves the effect of the intended instruction even under strong adversarial content. To "de-instructionalize" data tokens, DRIP introduces a data curation and training paradigm with a lightweight representation-editing module that edits embeddings of instruction-like tokens in the data section, enhancing security without harming utility. To ensure non-overwritability of instructions, DRIP adds a minimal residual module that reduces the ability of adversarial data to overwrite the original instruction. We evaluate DRIP on LLaMA 8B and Mistral 7B against StruQ, SecAlign, ISE, and PFT on three prompt-injection benchmarks (SEP, AlpacaFarm, and InjecAgent). DRIP improves role-separation score by 12-49\%, reduces attack success rate by over 66\% under adaptive attacks, and matches the utility of the undefended model, establishing a new state of the art for prompt-injection robustness.

Modern LLMs are trained to "think" primarily via explicit text generation, such as chain-of-thought (CoT), which defers reasoning to post-training and under-leverages pre-training data. We present and open-source Ouro, named after the recursive Ouroboros, a family of pre-trained Looped Language Models (LoopLM) that instead build reasoning into the pre-training phase through (i) iterative computation in latent space, (ii) an entropy-regularized objective for learned depth allocation, and (iii) scaling to 7.7T tokens. Ouro 1.4B and 2.6B models enjoy superior performance that match the results of up to 12B SOTA LLMs across a wide range of benchmarks. Through controlled experiments, we show this advantage stems not from increased knowledge capacity, but from superior knowledge manipulation capabilities. We also show that LoopLM yields reasoning traces more aligned with final outputs than explicit CoT. We hope our results show the potential of LoopLM as a novel scaling direction in the reasoning era. Our model is available here: http://ouro-llm.github.io.

Knowledge Base Question Answering (KBQA) aims to answer natural-language questions over a structured Knowledge Base (KB). Recent work improves KBQA by adopting an agentic reasoning paradigm, in which Large Language Models (LLMs) iteratively decompose a question, generate its corresponding logical queries, and interact with the KB to derive the answer. However, these methods typically fine-tune LLMs on reasoning trajectories synthesized via process supervision, which offers weak incentives for exploration and thus fails to strengthen the agentic reasoning ability. In this paper, we propose KnowCoder-A1, an LLM that can autonomously perform agentic reasoning on KBs to obtain answers. To incentivize autonomous exploration, KnowCoder-A1 trains the LLM under outcome-only supervision via a multi-stage curriculum reinforcement learning with an easy-to-hard curriculum. To establish foundational agentic capabilities, KnowCoder-A1 first fine-tunes the LLM on a small set of high-quality trajectories obtained through outcome-based rejection sampling. Then, to alleviate the reward sparsity inherent in outcome-only supervision, it applies multi-stage curriculum RL with reward schedules that progress from easy to hard. Trained with outcome-only supervision, KnowCoder-A1 exhibits powerful reasoning behaviors and consistently outperforms prior approaches across three mainstream datasets. Notably, on the zero-shot subset of GrailQA, KnowCoder-A1 achieves up to an 11.1% relative improvement while using only one-twelfth of the training data, demonstrating strong agentic reasoning capabilities.

Protected health information (PHI) de-identification is critical for enabling the safe reuse of clinical notes, yet evaluating and comparing PHI de-identification models typically depends on costly, small-scale expert annotations. We present TEAM-PHI, a multi-agent evaluation and selection framework that uses large language models (LLMs) to automatically measure de-identification quality and select the best-performing model without heavy reliance on gold labels. TEAM-PHI deploys multiple Evaluation Agents, each independently judging the correctness of PHI extractions and outputting structured metrics. Their results are then consolidated through an LLM-based majority voting mechanism that integrates diverse evaluator perspectives into a single, stable, and reproducible ranking. Experiments on a real-world clinical note corpus demonstrate that TEAM-PHI produces consistent and accurate rankings: despite variation across individual evaluators, LLM-based voting reliably converges on the same top-performing systems. Further comparison with ground-truth annotations and human evaluation confirms that the framework's automated rankings closely match supervised evaluation. By combining independent evaluation agents with LLM majority voting, TEAM-PHI offers a practical, secure, and cost-effective solution for automatic evaluation and best-model selection in PHI de-identification, even when ground-truth labels are limited.

Financial markets face growing threats from misinformation that can trigger billions in losses in minutes. Most existing approaches lack transparency in their decision-making and provide limited attribution to credible sources. We introduce FinVet, a novel multi-agent framework that integrates two Retrieval-Augmented Generation (RAG) pipelines with external fact-checking through a confidence-weighted voting mechanism. FinVet employs adaptive three-tier processing that dynamically adjusts verification strategies based on retrieval confidence, from direct metadata extraction to hybrid reasoning to full model-based analysis. Unlike existing methods, FinVet provides evidence-backed verdicts, source attribution, confidence scores, and explicit uncertainty flags when evidence is insufficient. Experimental evaluation on the FinFact dataset shows that FinVet achieves an F1 score of 0.85, which is a 10.4% improvement over the best individual pipeline (fact-check pipeline) and 37% improvement over standalone RAG approaches.

Reliable behavior control is central to deploying Large Language Models (LLMs) on the web. Activation steering offers a tuning-free route to align attributes (e.g., truthfulness) that ensure trustworthy generation. Prevailing approaches rely on coarse heuristics and lack a principled account of where to steer and how strongly to intervene. To this end, we propose P osition-wise I njection with eX act E stimated L evels (PIXEL), a position-wise activation steering framework that, in contrast to prior work, learns a property-aligned subspace from dual views (tail-averaged and end-token) and selects intervention strength via a constrained geometric objective with a closed-form solution, thereby adapting to token-level sensitivity without global hyperparameter tuning. PIXEL further performs sample-level orthogonal residual calibration to refine the global attribute direction and employs a lightweight position-scanning routine to identify receptive injection sites. We additionally provide representation-level guarantees for the minimal-intervention rule, supporting reliable alignment. Across diverse models and evaluation paradigms, PIXEL consistently improves attribute alignment while preserving model general capabilities, offering a practical and principled method for LLMs' controllable generation. To meet this need, a growing body of work has focused on post-training control mechanisms, which aim to adjust model behavior without retraining the entire model.

The BLP-2025 Task 1A requires Bengali hate speech classification into six categories. Traditional supervised approaches need extensive labeled datasets that are expensive for low-resource languages. We developed PromptGuard, a few-shot framework combining chi-square statistical analysis for keyword extraction with adaptive majority voting for decision-making. We explore statistical keyword selection versus random approaches and adaptive voting mechanisms that extend classification based on consensus quality. Chi-square keywords provide consistent improvements across categories, while adaptive voting benefits ambiguous cases requiring extended classification rounds. PromptGuard achieves a micro-F1 of 67.61, outperforming n-gram baselines (60.75) and random approaches (14.65). Ablation studies confirm chi-square-based keywords show the most consistent impact across all categories.

Abstract--Driven by the rapid advancements of Large Language Models (LLMs), LLM-based agents have emerged as powerful intelligent systems capable of human-like cognition, reasoning, and interaction. These agents are increasingly being deployed across diverse real-world applications, including student education, scientific research, and financial analysis. However, despite their remarkable potential, LLM-based agents remain vulnerable to hallucination issues, which can result in erroneous task execution and undermine the reliability of the overall system design. Addressing this critical challenge requires a deep understanding and a systematic consolidation of recent advances on LLM-based agents. T o this end, we present the first comprehensive survey of hallucinations in LLM-based agents. By carefully analyzing the complete workflow of agents, we propose a new taxonomy that identifies different types of agent hallucinations occurring at different stages. Furthermore, we conduct an in-depth examination of eighteen triggering causes underlying the emergence of agent hallucinations. Through a detailed review of a large number of existing studies, we summarize approaches for hallucination mitigation and detection, and highlight promising directions for future research. We hope this survey will inspire further efforts toward addressing hallucinations in LLM-based agents, ultimately contributing to the development of more robust and reliable agent systems. Cao, K. Chen, S. Hu, and L. Guo are with Institute of Information Engineering, Chinese Academy of Sciences, School of Cyber Security, University of Chinese Academy of Sciences, Beijing, China. K. Wang is with Nanyang Technological University, Singapore. Cao is with Institute of Automation, Chinese Academy of Sciences, Beijing, China. Q. Wang is with Hong Kong University of Science and Technology, Hong Kong, China. L. Zou is with School of Cyber Science and Engineering, Wuhan University, Wuhan, China. X. Chen is with Gaoling School of Artificial Intelligence, Renmin University of China, Beijing, China. C. Zhou is with Academy of Mathematics and Systems Science, Chinese Academy of Sciences, Beijing, China. J. Wu is with School of Computing, Faculty of Science and Engineering, Macquarie University, Sydney, Australia. Zhang is with the Cyberspace Institute of Advanced Technology, Guangzhou University, Guangzhou, China. Q. Wen is with Squirrel Ai Learning, Bellevue, USA. S. Pan is with School of Information and Communication Technology, Griffith University, Gold Coast, Australia. B. Wang is with Xiaomi Company, Beijing, China.