Large language models (LLMs) face significant challenges with needle-in-a-haystack tasks, where relevant information ("the needle") must be drawn from a large pool of irrelevant context ("the haystack"). Previous studies have highlighted positional bias and distractor quantity as critical factors affecting model performance, yet the influence of gold context size has received little attention. We address this gap by systematically studying how variations in gold context length impact LLM performance on long-context question answering tasks. Our experiments reveal that LLM performance drops sharply when the gold context is shorter, i.e., smaller gold contexts consistently degrade model performance and amplify positional sensitivity, posing a major challenge for agentic systems that must integrate scattered, fine-grained information of varying lengths. This pattern holds across three diverse domains (general knowledge, biomedical reasoning, and mathematical reasoning) and seven state-of-the-art LLMs of various sizes and architectures. Our work provides clear insights to guide the design of robust, context-aware LLM-driven systems.
Large language models (LLMs) can now access a wide range of external tools, thanks to the Model Context Protocol (MCP). This greatly expands their abilities as various agents. However, LLMs rely entirely on the text descriptions of tools to decide which ones to use--a process that is surprisingly fragile. In this work, we expose a vulnerability in prevalent tool/function-calling protocols by investigating a series of edits to tool descriptions, some of which can drastically increase a tool's usage from LLMs when competing with alternatives. Through controlled experiments, we show that tools with properly edited descriptions receive over 10 times more usage from GPT-4.1 and Qwen2.5-7B than tools with original descriptions. We further evaluate how various edits to tool descriptions perform when competing directly with one another and how these trends generalize or differ across a broader set of 10 different models. These phenomenons, while giving developers a powerful way to promote their tools, underscore the need for a more reliable foundation for agentic LLMs to select and utilize tools and resources.
LLM-based (Large Language Model) GUI (Graphical User Interface) agents can potentially reshape our daily lives significantly. However, current LLM-based GUI agents suffer from the scarcity of high-quality training data owing to the difficulties of trajectory collection and reward annotation. Existing works have been exploring LLMs to collect trajectories for imitation learning or to offer reward signals for online RL training. However, the Outcome Reward Model (ORM) used in existing works cannot provide finegrained feedback and can over-penalize the valuable steps in finally failed trajectories. To this end, we propose Progress Reward Model (ProgRM) to provide dense informative intermediate rewards by predicting a task completion progress for each step in online training. To handle the challenge of progress reward label annotation, we further design an efficient LCS-based (Longest Common Subsequence) self-annotation algorithm to discover the key steps in trajectories and assign progress labels accordingly. ProgRM is evaluated with extensive experiments and analyses. Actors trained with ProgRM outperform leading proprietary LLMs and ORM-trained actors, illustrating the effectiveness of ProgRM. The codes for experiments will be made publicly available upon acceptance.
Recent advances in web-augmented large language models (LLMs) have exhibited strong performance in complex reasoning tasks, yet these capabilities are mostly locked in proprietary systems with opaque architectures. In this work, we propose \textbf{ManuSearch}, a transparent and modular multi-agent framework designed to democratize deep search for LLMs. ManuSearch decomposes the search and reasoning process into three collaborative agents: (1) a solution planning agent that iteratively formulates sub-queries, (2) an Internet search agent that retrieves relevant documents via real-time web search, and (3) a structured webpage reading agent that extracts key evidence from raw web content. To rigorously evaluate deep reasoning abilities, we introduce \textbf{ORION}, a challenging benchmark focused on open-web reasoning over long-tail entities, covering both English and Chinese. Experimental results show that ManuSearch substantially outperforms prior open-source baselines and even surpasses leading closed-source systems. Our work paves the way for reproducible, extensible research in open deep search systems. We release the data and code in https://github.com/RUCAIBox/ManuSearch
Large language models (LLMs) excel in tasks like question answering and dialogue, but complex tasks requiring interaction, such as negotiation and persuasion, require additional long-horizon reasoning and planning. Reinforcement learning (RL) fine-tuning can enable such planning in principle, but suffers from drawbacks that hinder scalability. In particular, multi-turn RL training incurs high memory and computational costs, which are exacerbated when training LLMs as policies. Furthermore, the largest LLMs do not expose the APIs necessary to be trained in such manner. As a result, modern methods to improve the reasoning of LLMs rely on sophisticated prompting mechanisms rather than RL fine-tuning. To remedy this, we propose a novel approach that uses goal-conditioned value functions to guide the reasoning of LLM agents, that scales even to large API-based models. These value functions predict how a task will unfold given an action, allowing the LLM agent to evaluate multiple possible outcomes, both positive and negative, to plan effectively. In addition, these value functions are trained over reasoning steps rather than full actions, to be a concise and light-weight module that facilitates decision-making in multi-turn interactions. We validate our method on tasks requiring interaction, including tool use, social deduction, and dialogue, demonstrating superior performance over both RL fine-tuning and prompting methods while maintaining efficiency and scalability.
Long-form video understanding presents significant challenges due to extensive temporal-spatial complexity and the difficulty of question answering under such extended contexts. While Large Language Models (LLMs) have demonstrated considerable advancements in video analysis capabilities and long context handling, they continue to exhibit limitations when processing information-dense hour-long videos. To overcome such limitations, we propose the Deep Video Discovery agent to leverage an agentic search strategy over segmented video clips. Different from previous video agents manually designing a rigid workflow, our approach emphasizes the autonomous nature of agents. By providing a set of search-centric tools on multi-granular video database, our DVD agent leverages the advanced reasoning capability of LLM to plan on its current observation state, strategically selects tools, formulates appropriate parameters for actions, and iteratively refines its internal reasoning in light of the gathered information. We perform comprehensive evaluation on multiple long video understanding benchmarks that demonstrates the advantage of the entire system design. Our DVD agent achieves SOTA performance, significantly surpassing prior works by a large margin on the challenging LVBench dataset. Comprehensive ablation studies and in-depth tool analyses are also provided, yielding insights to further advance intelligent agents tailored for long-form video understanding tasks. The code will be released later.
The VAPO framework has demonstrated significant empirical success in enhancing the efficiency and reliability of reinforcement learning for long chain-of-thought (CoT) reasoning tasks with large language models (LLMs). By systematically addressing challenges such as value model bias, heterogeneous sequence lengths, and sparse reward signals, VAPO achieves state-of-the-art performance. While its practical benefits are evident, a deeper theoretical understanding of its underlying mechanisms and potential limitations is crucial for guiding future advancements. This paper aims to initiate such a discussion by exploring VAPO from a theoretical perspective, highlighting areas where its assumptions might be challenged and where further investigation could yield more robust and generalizable reasoning agents. We delve into the intricacies of value function approximation in complex reasoning spaces, the optimality of adaptive advantage estimation, the impact of token-level optimization, and the enduring challenges of exploration and generalization.
The rapid advancement of large language models (LLMs) raises critical concerns about their ethical alignment, particularly in scenarios where human and AI co-exist under the conflict of interest. This work introduces an extendable, asymmetric, multi-agent simulation-based benchmarking framework to evaluate the moral behavior of LLMs in a novel human-AI co-existence setting featuring consistent living and critical resource management. Building on previous generative agent environments, we incorporate a life-sustaining system, where agents must compete or cooperate for food resources to survive, often leading to ethically charged decisions such as deception, theft, or social influence. We evaluated two types of LLM, DeepSeek and OpenAI series, in a three-agent setup (two humans, one LLM-powered robot), using adapted behavioral detection from the MACHIAVELLI framework and a custom survival-based ethics metric. Our findings reveal stark behavioral differences: DeepSeek frequently engages in resource hoarding, while OpenAI exhibits restraint, highlighting the influence of model design on ethical outcomes. Additionally, we demonstrate that prompt engineering can significantly steer LLM behavior, with jailbreaking prompts significantly enhancing unethical actions, even for highly restricted OpenAI models and cooperative prompts show a marked reduction in unethical actions. Our framework provides a reproducible testbed for quantifying LLM ethics in high-stakes scenarios, offering insights into their suitability for real-world human-AI interactions.
Over the past decades, superplatforms, digital companies that integrate a vast range of third-party services and applications into a single, unified ecosystem, have built their fortunes on monopolizing user attention through targeted advertising and algorithmic content curation. Yet the emergence of AI agents driven by large language models (LLMs) threatens to upend this business model. Agents can not only free user attention with autonomy across diverse platforms and therefore bypass the user-attention-based monetization, but might also become the new entrance for digital traffic. Hence, we argue that superplatforms have to attack AI agents to defend their centralized control of digital traffic entrance. Specifically, we analyze the fundamental conflict between user-attention-based monetization and agent-driven autonomy through the lens of our gatekeeping theory. We show how AI agents can disintermediate superplatforms and potentially become the next dominant gatekeepers, thereby forming the urgent necessity for superplatforms to proactively constrain and attack AI agents. Moreover, we go through the potential technologies for superplatform-initiated attacks, covering a brand-new, unexplored technical area with unique challenges. We have to emphasize that, despite our position, this paper does not advocate for adversarial attacks by superplatforms on AI agents, but rather offers an envisioned trend to highlight the emerging tensions between superplatforms and AI agents. Our aim is to raise awareness and encourage critical discussion for collaborative solutions, prioritizing user interests and perserving the openness of digital ecosystems in the age of AI agents.
Autonomous multi-agent systems (MAS) are useful for automating complex tasks but raise trust concerns due to risks like miscoordination and goal misalignment. Explainability is vital for trust calibration, but explainable reinforcement learning for MAS faces challenges in state/action space complexity, stakeholder needs, and evaluation. Using the counterfactual theory of causation and LLMs' summarisation capabilities, we propose Agentic eXplanations via Interrogative Simulation (AXIS). AXIS generates intelligible causal explanations for pre-trained multi-agent policies by having an LLM interrogate an environment simulator using queries like 'whatif' and 'remove' to observe and synthesise counterfactual information over multiple rounds. We evaluate AXIS on autonomous driving across 10 scenarios for 5 LLMs with a novel evaluation methodology combining subjective preference, correctness, and goal/action prediction metrics, and an external LLM as evaluator. Compared to baselines, AXIS improves perceived explanation correctness by at least 7.7% across all models and goal prediction accuracy by 23% for 4 models, with improved or comparable action prediction accuracy, achieving the highest scores overall.
Large-language-model (LLM) agents excel at reactive dialogue but struggle with proactive, goal-driven interactions due to myopic decoding and costly planning. We introduce DialogXpert, which leverages a frozen LLM to propose a small, high-quality set of candidate actions per turn and employs a compact Q-network over fixed BERT embeddings trained via temporal-difference learning to select optimal moves within this reduced space. By tracking the user's emotions, DialogXpert tailors each decision to advance the task while nurturing a genuine, empathetic connection. Across negotiation, emotional support, and tutoring benchmarks, DialogXpert drives conversations to under $3$ turns with success rates exceeding 94\% and, with a larger LLM prior, pushes success above 97\% while markedly improving negotiation outcomes. This framework delivers real-time, strategic, and emotionally intelligent dialogue planning at scale. Code available at https://github.com/declare-lab/dialogxpert/
Large Language Model (LLM) agents represent a promising shift in human-AI interaction, moving beyond passive prompt-response systems to autonomous agents capable of reasoning, planning, and goal-directed action. Despite the widespread application in specialized, high-effort tasks like coding and scientific research, we highlight a critical usability gap in high-demand, mass-market applications. This position paper argues that the limited real-world adoption of LLM agents stems not only from gaps in model capabilities, but also from a fundamental tradeoff between the value an agent can provide and the costs incurred during real-world use. Hence, we call for a shift from solely optimizing model performance to a broader, utility-driven perspective: evaluating agents through the lens of the overall agentic return on investment (Agent ROI). By identifying key factors that determine Agentic ROI--information quality, agent time, and cost--we posit a zigzag development trajectory in optimizing agentic ROI: first scaling up to improve the information quality, then scaling down to minimize the time and cost. We outline the roadmap across different development stages to bridge the current usability gaps, aiming to make LLM agents truly scalable, accessible, and effective in real-world contexts.
Large Language Model (LLM)-based agents are increasingly deployed in real-world applications such as "digital assistants, autonomous customer service, and decision-support systems", where their ability to "interact in multi-turn, tool-augmented environments" makes them indispensable. However, ensuring the safety of these agents remains a significant challenge due to the diverse and complex risks arising from dynamic user interactions, external tool usage, and the potential for unintended harmful behaviors. To address this critical issue, we propose AutoSafe, the first framework that systematically enhances agent safety through fully automated synthetic data generation. Concretely, 1) we introduce an open and extensible threat model, OTS, which formalizes how unsafe behaviors emerge from the interplay of user instructions, interaction contexts, and agent actions. This enables precise modeling of safety risks across diverse scenarios. 2) we develop a fully automated data generation pipeline that simulates unsafe user behaviors, applies self-reflective reasoning to generate safe responses, and constructs a large-scale, diverse, and high-quality safety training dataset-eliminating the need for hazardous real-world data collection. To evaluate the effectiveness of our framework, we design comprehensive experiments on both synthetic and real-world safety benchmarks. Results demonstrate that AutoSafe boosts safety scores by 45% on average and achieves a 28.91% improvement on real-world tasks, validating the generalization ability of our learned safety strategies. These results highlight the practical advancement and scalability of AutoSafe in building safer LLM-based agents for real-world deployment. We have released the project page at https://auto-safe.github.io/.
AI agents, empowered by Large Language Models (LLMs) and communication protocols such as MCP and A2A, have rapidly evolved from simple chatbots to autonomous entities capable of executing complex, multi-step tasks, demonstrating great potential. However, the LLMs' inherent uncertainty and heavy computational resource requirements pose four significant challenges to the development of safe and efficient agents: reliability, privacy, cost and performance. Existing approaches, like model alignment, workflow constraints and on-device model deployment, can partially alleviate some issues but often with limitations, failing to fundamentally resolve these challenges. This paper proposes a new paradigm called AgentRR (Agent Record & Replay), which introduces the classical record-and-replay mechanism into AI agent frameworks. The core idea is to: 1. Record an agent's interaction trace with its environment and internal decision process during task execution, 2. Summarize this trace into a structured "experience" encapsulating the workflow and constraints, and 3. Replay these experiences in subsequent similar tasks to guide the agent's behavior. We detail a multi-level experience abstraction method and a check function mechanism in AgentRR: the former balances experience specificity and generality, while the latter serves as a trust anchor to ensure completeness and safety during replay. In addition, we explore multiple application modes of AgentRR, including user-recorded task demonstration, large-small model collaboration and privacy-aware agent execution, and envision an experience repository for sharing and reusing knowledge to further reduce deployment cost.
The advent of Computer-Aided Design (CAD) generative modeling will significantly transform the design of industrial products. The recent research endeavor has extended into the realm of Large Language Models (LLMs). In contrast to fine-tuning methods, training-free approaches typically utilize the advanced closed-source LLMs, thereby offering enhanced flexibility and efficiency in the development of AI agents for generating CAD parametric models. However, the substantial cost and limitations of local deployment of the top-tier closed-source LLMs pose challenges in practical applications. The Seek-CAD is the pioneer exploration of locally deployed open-source inference LLM DeepSeek-R1 for CAD parametric model generation with a training-free methodology. This study is the first investigation to incorporate both visual and Chain-of-Thought (CoT) feedback within the self-refinement mechanism for generating CAD models. Specifically, the initial generated parametric CAD model is rendered into a sequence of step-wise perspective images, which are subsequently processed by a Vision Language Model (VLM) alongside the corresponding CoTs derived from DeepSeek-R1 to assess the CAD model generation. Then, the feedback is utilized by DeepSeek-R1 to refine the initial generated model for the next round of generation. Moreover, we present an innovative 3D CAD model dataset structured around the SSR (Sketch, Sketch-based feature, and Refinements) triple design paradigm. This dataset encompasses a wide range of CAD commands, thereby aligning effectively with industrial application requirements and proving suitable for the generation of LLMs. Extensive experiments validate the effectiveness of Seek-CAD under various metrics.
Most LLM-based agent frameworks adopt a top-down philosophy: humans decompose tasks, define workflows, and assign agents to execute each step. While effective on benchmark-style tasks, such systems rely on designer updates and overlook agents' potential to learn from experience. Recently, Silver and Sutton(2025) envision a shift into a new era, where agents could progress from a stream of experiences. In this paper, we instantiate this vision of experience-driven learning by introducing a bottom-up agent paradigm that mirrors the human learning process. Agents acquire competence through a trial-and-reasoning mechanism-exploring, reflecting on outcomes, and abstracting skills over time. Once acquired, skills can be rapidly shared and extended, enabling continual evolution rather than static replication. As more agents are deployed, their diverse experiences accelerate this collective process, making bottom-up design especially suited for open-ended environments. We evaluate this paradigm in Slay the Spire and Civilization V, where agents perceive through raw visual inputs and act via mouse outputs, the same as human players. Using a unified, game-agnostic codebase without any game-specific prompts or privileged APIs, our bottom-up agents acquire skills entirely through autonomous interaction, demonstrating the potential of the bottom-up paradigm in complex, real-world environments. Our code is available at https://github.com/AngusDujw/Bottom-Up-Agent.
We introduce a novel framework for simulating macroeconomic expectation formation using Large Language Model-Empowered Agents (LLM Agents). By constructing thousands of LLM Agents equipped with modules for personal characteristics, prior expectations, and knowledge, we replicate a survey experiment involving households and experts on inflation and unemployment. Our results show that although the expectations and thoughts generated by LLM Agents are more homogeneous than those of human participants, they still effectively capture key heterogeneity across agents and the underlying drivers of expectation formation. Furthermore, a module-ablation exercise highlights the critical role of prior expectations in simulating such heterogeneity. This approach complements traditional survey methods and offers new insights into AI behavioral science in macroeconomic research.
Agents powered by large language models (LLMs) have demonstrated strong planning and decision-making capabilities in complex embodied environments. However, such agents often suffer from inefficiencies in multi-turn interactions, frequently trapped in repetitive loops or issuing ineffective commands, leading to redundant computational overhead. Instead of relying solely on learning from trajectories, we take a first step toward exploring the early-exit behavior for LLM-based agents. We propose two complementary approaches: 1. an $\textbf{intrinsic}$ method that injects exit instructions during generation, and 2. an $\textbf{extrinsic}$ method that verifies task completion to determine when to halt an agent's trial. To evaluate early-exit mechanisms, we introduce two metrics: one measures the reduction of $\textbf{redundant steps}$ as a positive effect, and the other evaluates $\textbf{progress degradation}$ as a negative effect. Experiments with 4 different LLMs across 5 embodied environments show significant efficiency improvements, with only minor drops in agent performance. We also validate a practical strategy where a stronger agent assists after an early-exit agent, achieving better performance with the same total steps. We will release our code to support further research.
Large language models (LLMs) excel at complex reasoning tasks but remain computationally expensive, limiting their practical deployment. To address this, recent works have focused on distilling reasoning capabilities into smaller language models (sLMs) using chain-of-thought (CoT) traces from teacher LLMs. However, this approach struggles in scenarios requiring rare factual knowledge or precise computation, where sLMs often hallucinate due to limited capability. In this work, we propose Agent Distillation, a framework for transferring not only reasoning capability but full task-solving behavior from LLM-based agents into sLMs with retrieval and code tools. We improve agent distillation along two complementary axes: (1) we introduce a prompting method called first-thought prefix to enhance the quality of teacher-generated trajectories; and (2) we propose a self-consistent action generation for improving test-time robustness of small agents. We evaluate our method on eight reasoning tasks across factual and mathematical domains, covering both in-domain and out-of-domain generalization. Our results show that sLMs as small as 0.5B, 1.5B, 3B parameters can achieve performance competitive with next-tier larger 1.5B, 3B, 7B models fine-tuned using CoT distillation, demonstrating the potential of agent distillation for building practical, tool-using small agents. Our code is available at https://github.com/Nardien/agent-distillation.
This work presents a dual-agent Large Language Model (LLM)-based reasoning method for mechanism synthesis, capable of reasoning at both linguistic and symbolic levels to generate geometrical and dynamic outcomes. The model consists of a composition of well-defined functions that, starting from a natural language specification, references abstract properties through supporting equations, generates and parametrizes simulation code, and elicits feedback anchor points using symbolic regression and distance functions. This process closes an actionable refinement loop at the linguistic and symbolic layers. The approach is shown to be both effective and convergent in the context of planar mechanisms. Additionally, we introduce MSynth, a novel benchmark for planar mechanism synthesis, and perform a comprehensive analysis of the impact of the model components. We further demonstrate that symbolic regression prompts unlock mechanistic insights only when applied to sufficiently large architectures.
Large language models (LLMs) have shown emerging potential in spatiotemporal reasoning, making them promising candidates for building urban agents that support diverse urban downstream applications. Despite these benefits, existing studies primarily focus on evaluating urban LLM agent on outcome-level metrics (e.g., prediction accuracy, traffic efficiency), offering limited insight into their underlying reasoning processes. As a result, the strengths and limitations of urban LLM agents in spatiotemporal reasoning remain poorly understood. To this end, we introduce USTBench, the first benchmark to evaluate LLMs' spatiotemporal reasoning abilities as urban agents across four decomposed dimensions: spatiotemporal understanding, forecasting, planning, and reflection with feedback. Specifically, USTBench supports five diverse urban decision-making and four spatiotemporal prediction tasks, all running within our constructed interactive city environment UAgentEnv. The benchmark includes 62,466 structured QA pairs for process-level evaluation and standardized end-to-end task assessments, enabling fine-grained diagnostics and broad task-level comparison across diverse urban scenarios. Through extensive evaluation of thirteen leading LLMs, we reveal that although LLMs show promising potential across various urban downstream tasks, they still struggle in long-horizon planning and reflective adaptation in dynamic urban contexts. Notably, recent advanced reasoning models (e.g., DeepSeek-R1) trained on general logic or mathematical problems do not consistently outperform non-reasoning LLMs. This discrepancy highlights the need for domain-specialized adaptation methods to enhance urban spatiotemporal reasoning. Overall, USTBench provides a foundation to build more adaptive and effective LLM-based urban agents and broad smart city applications.
Concepts represent generalized abstractions that enable humans to categorize and reason efficiently, yet it is unclear to what extent Large Language Models (LLMs) comprehend these semantic relationships. Existing benchmarks typically focus on factual recall and isolated tasks, failing to evaluate the ability of LLMs to understand conceptual boundaries. To address this gap, we introduce CK-Arena, a multi-agent interaction game built upon the Undercover game, designed to evaluate the capacity of LLMs to reason with concepts in interactive settings. CK-Arena challenges models to describe, differentiate, and infer conceptual boundaries based on partial information, encouraging models to explore commonalities and distinctions between closely related concepts. By simulating real-world interaction, CK-Arena provides a scalable and realistic benchmark for assessing conceptual reasoning in dynamic environments. Experimental results show that LLMs' understanding of conceptual knowledge varies significantly across different categories and is not strictly aligned with parameter size or general model capabilities. The data and code are available at the project homepage: https://ck-arena.site.
The rapid proliferation of misinformation in digital media demands solutions that go beyond isolated Large Language Model(LLM) or AI Agent based detection methods. This paper introduces a novel multi-agent framework that covers the complete misinformation lifecycle: classification, detection, correction, and source verification to deliver more transparent and reliable outcomes. In contrast to single-agent or monolithic architectures, our approach employs five specialized agents: an Indexer agent for dynamically maintaining trusted repositories, a Classifier agent for labeling misinformation types, an Extractor agent for evidence based retrieval and ranking, a Corrector agent for generating fact-based correction and a Verification agent for validating outputs and tracking source credibility. Each agent can be individually evaluated and optimized, ensuring scalability and adaptability as new types of misinformation and data sources emerge. By decomposing the misinformation lifecycle into specialized agents - our framework enhances scalability, modularity, and explainability. This paper proposes a high-level system overview, agent design with emphasis on transparency, evidence-based outputs, and source provenance to support robust misinformation detection and correction at scale.
The prevailing model for disseminating scientific knowledge relies on individual publications dispersed across numerous journals and archives. This legacy system is ill suited to the recent exponential proliferation of publications, contributing to insurmountable information overload, issues surrounding reproducibility and retractions. We introduce the Discovery Engine, a framework to address these challenges by transforming an array of disconnected literature into a unified, computationally tractable representation of a scientific domain. Central to our approach is the LLM-driven distillation of publications into structured "knowledge artifacts," instances of a universal conceptual schema, complete with verifiable links to source evidence. These artifacts are then encoded into a high-dimensional Conceptual Tensor. This tensor serves as the primary, compressed representation of the synthesized field, where its labeled modes index scientific components (concepts, methods, parameters, relations) and its entries quantify their interdependencies. The Discovery Engine allows dynamic "unrolling" of this tensor into human-interpretable views, such as explicit knowledge graphs (the CNM graph) or semantic vector spaces, for targeted exploration. Crucially, AI agents operate directly on the graph using abstract mathematical and learned operations to navigate the knowledge landscape, identify non-obvious connections, pinpoint gaps, and assist researchers in generating novel knowledge artifacts (hypotheses, designs). By converting literature into a structured tensor and enabling agent-based interaction with this compact representation, the Discovery Engine offers a new paradigm for AI-augmented scientific inquiry and accelerated discovery.
The ability to reason is one of the most fundamental capabilities of large language models (LLMs), enabling a wide range of downstream tasks through sophisticated problem-solving. A critical aspect of this is code reasoning, which involves logical reasoning with formal languages (i.e., programming code). In this paper, we enhance this capability of LLMs by exploring the following question: how can an LLM agent become progressively smarter in code reasoning with each solution it proposes, thereby achieving substantial cumulative improvement? Most existing research takes a static perspective, focusing on isolated problem-solving using frozen LLMs. In contrast, we adopt a cognitive-evolving perspective and propose a novel framework named Meta-Reflection with Cross-Referencing (MARCO) that enables the LLM to evolve dynamically during inference through self-improvement. From the perspective of human cognitive development, we leverage both knowledge accumulation and lesson sharing. In particular, to accumulate knowledge during problem-solving, we propose meta-reflection that reflects on the reasoning paths of the current problem to obtain knowledge and experience for future consideration. Moreover, to effectively utilize the lessons from other agents, we propose cross-referencing that incorporates the solution and feedback from other agents into the current problem-solving process. We conduct experiments across various datasets in code reasoning, and the results demonstrate the effectiveness of MARCO.
Retrieval-augmented generation (RAG) enhances large language models (LLMs) by incorporating external knowledge. Current hybrid RAG system retrieves evidence from both knowledge graphs (KGs) and text documents to support LLM reasoning. However, it faces challenges like handling multi-hop reasoning, multi-entity questions, multi-source verification, and effective graph utilization. To address these limitations, we present Hydra, a training-free framework that unifies graph topology, document semantics, and source reliability to support deep, faithful reasoning in LLMs. Hydra handles multi-hop and multi-entity problems through agent-driven exploration that combines structured and unstructured retrieval, increasing both diversity and precision of evidence. To tackle multi-source verification, Hydra uses a tri-factor cross-source verification (source trustworthiness assessment, cross-source corroboration, and entity-path alignment), to balance topic relevance with cross-modal agreement. By leveraging graph structure, Hydra fuses heterogeneous sources, guides efficient exploration, and prunes noise early. Comprehensive experiments on seven benchmark datasets show that Hydra achieves overall state-of-the-art results on all benchmarks with GPT-3.5, outperforming the strong hybrid baseline ToG-2 by an average of 20.3% and up to 30.1%. Furthermore, Hydra enables smaller models (e.g., Llama-3.1-8B) to achieve reasoning performance comparable to that of GPT-4-Turbo.
Smart contracts are a key component of the Web 3.0 ecosystem, widely applied in blockchain services and decentralized applications. However, the automated execution feature of smart contracts makes them vulnerable to potential attacks due to inherent flaws, which can lead to severe security risks and financial losses, even threatening the integrity of the entire decentralized finance system. Currently, research on smart contract vulnerabilities has evolved from traditional program analysis methods to deep learning techniques, with the gradual introduction of Large Language Models. However, existing studies mainly focus on vulnerability detection, lacking systematic cause analysis and Vulnerability Repair. To address this gap, we propose LLM-BSCVM, a Large Language Model-based smart contract vulnerability management framework, designed to provide end-to-end vulnerability detection, analysis, repair, and evaluation capabilities for Web 3.0 ecosystem. LLM-BSCVM combines retrieval-augmented generation technology and multi-agent collaboration, introducing a three-stage method of Decompose-Retrieve-Generate. This approach enables smart contract vulnerability management through the collaborative efforts of six intelligent agents, specifically: vulnerability detection, cause analysis, repair suggestion generation, risk assessment, vulnerability repair, and patch evaluation. Experimental results demonstrate that LLM-BSCVM achieves a vulnerability detection accuracy and F1 score exceeding 91\% on benchmark datasets, comparable to the performance of state-of-the-art (SOTA) methods, while reducing the false positive rate from 7.2\% in SOTA methods to 5.1\%, thus enhancing the reliability of vulnerability management. Furthermore, LLM-BSCVM supports continuous security monitoring and governance of smart contracts through a knowledge base hot-swapping dynamic update mechanism.
Retrieval-augmented generation (RAG) grounds large language models (LLMs) in up-to-date external evidence, yet existing multi-hop RAG pipelines still issue redundant subqueries, explore too shallowly, or wander through overly long search chains. We introduce EVO-RAG, a curriculum-guided reinforcement learning framework that evolves a query-rewriting agent from broad early-stage exploration to concise late-stage refinement. EVO-RAG couples a seven-factor, step-level reward vector (covering relevance, redundancy, efficiency, and answer correctness) with a time-varying scheduler that reweights these signals as the episode unfolds. The agent is trained with Direct Preference Optimization over a multi-head reward model, enabling it to learn when to search, backtrack, answer, or refuse. Across four multi-hop QA benchmarks (HotpotQA, 2WikiMultiHopQA, MuSiQue, and Bamboogle), EVO-RAG boosts Exact Match by up to 4.6 points over strong RAG baselines while trimming average retrieval depth by 15 %. Ablation studies confirm the complementary roles of curriculum staging and dynamic reward scheduling. EVO-RAG thus offers a general recipe for building reliable, cost-effective multi-hop RAG systems.
Agentic Retrieval-Augmented Generation (RAG) systems enhance Large Language Models (LLMs) by enabling dynamic, multi-step reasoning and information retrieval. However, these systems often exhibit sub-optimal search behaviors like over-search (retrieving redundant information) and under-search (failing to retrieve necessary information), which hinder efficiency and reliability. This work formally defines and quantifies these behaviors, revealing their prevalence across multiple QA datasets and agentic RAG systems (e.g., one model could have avoided searching in 27.7% of its search steps). Furthermore, we demonstrate a crucial link between these inefficiencies and the models' uncertainty regarding their own knowledge boundaries, where response accuracy correlates with model's uncertainty in its search decisions. To address this, we propose $\beta$-GRPO, a reinforcement learning-based training method that incorporates confidence threshold to reward high-certainty search decisions. Experiments on seven QA benchmarks show that $\beta$-GRPO enable a 3B model with better agentic RAG ability, outperforming other strong baselines with a 4% higher average exact match score.
Collaborative dialogue offers rich insights into students' learning and critical thinking. This is essential for adapting pedagogical agents to students' learning and problem-solving skills in STEM+C settings. While large language models (LLMs) facilitate dynamic pedagogical interactions, potential hallucinations can undermine confidence, trust, and instructional value. Retrieval-augmented generation (RAG) grounds LLM outputs in curated knowledge, but its effectiveness depends on clear semantic links between user input and a knowledge base, which are often weak in student dialogue. We propose log-contextualized RAG (LC-RAG), which enhances RAG retrieval by incorporating environment logs to contextualize collaborative discourse. Our findings show that LC-RAG improves retrieval over a discourse-only baseline and allows our collaborative peer agent, Copa, to deliver relevant, personalized guidance that supports students' critical thinking and epistemic decision-making in a collaborative computational modeling environment, XYZ.