We present \textsc{MineNPC-Task}, a user-authored benchmark and evaluation harness for testing memory-aware, mixed-initiative LLM agents in open-world \emph{Minecraft}. Rather than relying on synthetic prompts, tasks are elicited from formative and summative co-play with expert players, normalized into parametric templates with explicit preconditions and dependency structure, and paired with machine-checkable validators under a bounded-knowledge policy that forbids out-of-world shortcuts. The harness captures plan/act/memory events-including plan previews, targeted clarifications, memory reads and writes, precondition checks, and repair attempts and reports outcomes relative to the total number of attempted subtasks, derived from in-world evidence. As an initial snapshot, we instantiate the framework with GPT-4o and evaluate \textbf{216} subtasks across \textbf{8} experienced players. We observe recurring breakdown patterns in code execution, inventory/tool handling, referencing, and navigation, alongside recoveries supported by mixed-initiative clarifications and lightweight memory. Participants rated interaction quality and interface usability positively, while highlighting the need for stronger memory persistence across tasks. We release the complete task suite, validators, logs, and harness to support transparent, reproducible evaluation of future memory-aware embodied agents.
Large Language Models (LLMs) have shown remarkable capabilities in tool calling and tool usage, but suffer from hallucinations where they choose incorrect tools, provide malformed parameters and exhibit 'tool bypass' behavior by performing simulations and generating outputs instead of invoking specialized tools or external systems. This undermines the reliability of LLM based agents in production systems as it leads to inconsistent results, and bypasses security and audit controls. Such hallucinations in agent tool selection require early detection and error handling. Unlike existing hallucination detection methods that require multiple forward passes or external validation, we present a computationally efficient framework that detects tool-calling hallucinations in real-time by leveraging LLMs' internal representations during the same forward pass used for generation. We evaluate this approach on reasoning tasks across multiple domains, demonstrating strong detection performance (up to 86.4\% accuracy) while maintaining real-time inference capabilities with minimal computational overhead, particularly excelling at detecting parameter-level hallucinations and inappropriate tool selections, critical for reliable agent deployment.
Large language models (LLMs) have revolutionized text-based code automation, but their potential in graph-oriented engineering workflows remains under-explored. We introduce SimuAgent, an LLM-powered modeling and simulation agent tailored for Simulink. SimuAgent replaces verbose XML with a concise, dictionary-style Python representation, dramatically cutting token counts, improving interpretability, and enabling fast, in-process simulation. A lightweight plan-execute architecture, trained in two stages, equips the agent with both low-level tool skills and high-level design reasoning. To tackle sparse rewards in long-horizon tasks, we propose Reflection-GRPO (ReGRPO), which augments Group Relative Policy Optimization (GRPO) with self-reflection traces that supply rich intermediate feedback, accelerating convergence and boosting robustness. Experiments on SimuBench, our newly released benchmark comprising 5300 multi-domain modeling tasks, show that a Qwen2.5-7B model fine-tuned with SimuAgent converges faster and achieves higher modeling accuracy than standard RL baselines, and even surpasses GPT-4o when evaluated with few-shot prompting on the same benchmark. Ablations confirm that the two-stage curriculum and abstract-reconstruct data augmentation further enhance generalization. SimuAgent trains and runs entirely on-premise with modest hardware, delivering a privacy-preserving, cost-effective solution for industrial model-driven engineering. SimuAgent bridges the gap between LLMs and graphical modeling environments, offering a practical solution for AI-assisted engineering design in industrial settings.
Embodied question answering (EQA) in 3D environments often requires collecting context that is distributed across multiple viewpoints and partially occluded. However, most recent vision--language models (VLMs) are constrained to a fixed and finite set of input views, which limits their ability to acquire question-relevant context at inference time and hinders complex spatial reasoning. We propose Chain-of-View (CoV) prompting, a training-free, test-time reasoning framework that transforms a VLM into an active viewpoint reasoner through a coarse-to-fine exploration process. CoV first employs a View Selection agent to filter redundant frames and identify question-aligned anchor views. It then performs fine-grained view adjustment by interleaving iterative reasoning with discrete camera actions, obtaining new observations from the underlying 3D scene representation until sufficient context is gathered or a step budget is reached. We evaluate CoV on OpenEQA across four mainstream VLMs and obtain an average +11.56\% improvement in LLM-Match, with a maximum gain of +13.62\% on Qwen3-VL-Flash. CoV further exhibits test-time scaling: increasing the minimum action budget yields an additional +2.51\% average improvement, peaking at +3.73\% on Gemini-2.5-Flash. On ScanQA and SQA3D, CoV delivers strong performance (e.g., 116 CIDEr / 31.9 EM@1 on ScanQA and 51.1 EM@1 on SQA3D). Overall, these results suggest that question-aligned view selection coupled with open-view search is an effective, model-agnostic strategy for improving spatial reasoning in 3D EQA without additional training.
LLM-as-a-Judge has revolutionized AI evaluation by leveraging large language models for scalable assessments. However, as evaluands become increasingly complex, specialized, and multi-step, the reliability of LLM-as-a-Judge has become constrained by inherent biases, shallow single-pass reasoning, and the inability to verify assessments against real-world observations. This has catalyzed the transition to Agent-as-a-Judge, where agentic judges employ planning, tool-augmented verification, multi-agent collaboration, and persistent memory to enable more robust, verifiable, and nuanced evaluations. Despite the rapid proliferation of agentic evaluation systems, the field lacks a unified framework to navigate this shifting landscape. To bridge this gap, we present the first comprehensive survey tracing this evolution. Specifically, we identify key dimensions that characterize this paradigm shift and establish a developmental taxonomy. We organize core methodologies and survey applications across general and professional domains. Furthermore, we analyze frontier challenges and identify promising research directions, ultimately providing a clear roadmap for the next generation of agentic evaluation.
LLM-driven agentic applications increasingly automate complex, multi-step tasks, but serving them efficiently remains challenging due to heterogeneous components, dynamic and model-driven control flow, long-running state, and unpredictable latencies. Nalar is a ground-up agent-serving framework that cleanly separates workflow specification from execution while providing the runtime visibility and control needed for robust performance. Nalar preserves full Python expressiveness, using lightweight auto-generated stubs that turn agent and tool invocations into futures carrying dependency and context metadata. A managed state layer decouples logical state from physical placement, enabling safe reuse, migration, and consistent retry behavior. A two-level control architecture combines global policy computation with local event-driven enforcement to support adaptive routing, scheduling, and resource management across evolving workflows. Together, these mechanisms allow Nalar to deliver scalable, efficient, and policy-driven serving of heterogeneous agentic applications without burdening developers with orchestration logic. Across three agentic workloads, Nalar cuts tail latency by 34--74\%, achieves up to $2.9\times$ speedups, sustains 80 RPS where baselines fail, and scales to 130K futures with sub-500 ms control overhead.
As LLM-based agents are increasingly used in long-term interactions, cumulative memory is critical for enabling personalization and maintaining stylistic consistency. However, most existing systems adopt an ``all-or-nothing'' approach to memory usage: incorporating all relevant past information can lead to \textit{Memory Anchoring}, where the agent is trapped by past interactions, while excluding memory entirely results in under-utilization and the loss of important interaction history. We show that an agent's reliance on memory can be modeled as an explicit and user-controllable dimension. We first introduce a behavioral metric of memory dependence to quantify the influence of past interactions on current outputs. We then propose \textbf{Stee}rable \textbf{M}emory Agent, \texttt{SteeM}, a framework that allows users to dynamically regulate memory reliance, ranging from a fresh-start mode that promotes innovation to a high-fidelity mode that closely follows interaction history. Experiments across different scenarios demonstrate that our approach consistently outperforms conventional prompting and rigid memory masking strategies, yielding a more nuanced and effective control for personalized human-agent collaboration.
Large Language Models (LLMs) are now integral to numerous industries, increasingly serving as the core reasoning engine for autonomous agents that perform complex tasks through tool-use. While the development of Arabic-native LLMs is accelerating, the benchmarks for evaluating their capabilities lag behind, with most existing frameworks focusing on English. A critical and overlooked area is tool-calling, where the performance of models prompted in non-English languages like Arabic is poorly understood, especially since these models are often pretrained on predominantly English data. This paper addresses this critical gap by introducing the first dedicated benchmark for evaluating the tool-calling and agentic capabilities of LLMs in the Arabic language. Our work provides a standardized framework to measure the functional accuracy and robustness of models in Arabic agentic workflows. Our findings reveal a huge performance gap: when users interact in Arabic, tool-calling accuracy drops by an average of 5-10\%, regardless of whether the tool descriptions themselves are in Arabic or English. By shedding light on these critical challenges, this benchmark aims to foster the development of more reliable and linguistically equitable AI agents for Arabic-speaking users.
Deep Research (DR) Agents powered by advanced Large Language Models (LLMs) have fundamentally shifted the paradigm for completing complex research tasks. Yet, a comprehensive and live evaluation of their forecasting performance on real-world, research-oriented tasks in high-stakes domains (e.g., finance) remains underexplored. We introduce FinDeepForecast, the first live, end-to-end multi-agent system for automatically evaluating DR agents by continuously generating research-oriented financial forecasting tasks. This system is equipped with a dual-track taxonomy, enabling the dynamic generation of recurrent and non-recurrent forecasting tasks at both corporate and macro levels. With this system, we generate FinDeepForecastBench, a weekly evaluation benchmark over a ten-week horizon, encompassing 8 global economies and 1,314 listed companies, and evaluate 13 representative methods. Extensive experiments show that, while DR agents consistently outperform strong baselines, their performance still falls short of genuine forward-looking financial reasoning. We expect the proposed FinDeepForecast system to consistently facilitate future advancements of DR agents in research-oriented financial forecasting tasks. The benchmark and leaderboard are publicly available on the OpenFinArena Platform.
Self-destructive behaviors are linked to complex psychological states and can be challenging to diagnose. These behaviors may be even harder to identify within subcultural groups due to their unique expressions. As large language models (LLMs) are applied across various fields, some researchers have begun exploring their application for detecting self-destructive behaviors. Motivated by this, we investigate self-destructive behavior detection within subcultures using current LLM-based methods. However, these methods have two main challenges: (1) Knowledge Lag: Subcultural slang evolves rapidly, faster than LLMs' training cycles; and (2) Semantic Misalignment: it is challenging to grasp the specific and nuanced expressions unique to subcultures. To address these issues, we proposed Subcultural Alignment Solver (SAS), a multi-agent framework that incorporates automatic retrieval and subculture alignment, significantly enhancing the performance of LLMs in detecting self-destructive behavior. Our experimental results show that SAS outperforms the current advanced multi-agent framework OWL. Notably, it competes well with fine-tuned LLMs. We hope that SAS will advance the field of self-destructive behavior detection in subcultural contexts and serve as a valuable resource for future researchers.
Large language model (LLM)-based search agents have proven promising for addressing knowledge-intensive problems by incorporating information retrieval capabilities. Existing works largely focus on optimizing the reasoning paradigms of search agents, yet the quality of intermediate search queries during reasoning remains overlooked. As a result, the generated queries often remain inaccurate, leading to unexpected retrieval results and ultimately limiting search agents' overall effectiveness. To mitigate this issue, we introduce SmartSearch, a framework built upon two key mechanisms: (1) Process rewards, which provide fine-grained supervision for the quality of each intermediate search query through Dual-Level Credit Assessment. (2) Query refinement, which promotes the optimization of query generation by selectively refining low-quality search queries and regenerating subsequent search rounds based on these refinements. To enable the search agent to progressively internalize the ability to improve query quality under the guidance of process rewards, we design a three-stage curriculum learning framework. This framework guides the agent through a progression from imitation, to alignment, and ultimately to generalization. Experimental results show that SmartSearch consistently surpasses existing baselines, and additional quantitative analyses further confirm its significant gains in both search efficiency and query quality. The code is available at https://github.com/MYVAE/SmartSearch.
Synthesizing informative commercial reports from massive and noisy web sources is critical for high-stakes business decisions. Although current deep research agents achieve notable progress, their reports still remain limited in terms of quality, reliability, and coverage. In this work, we propose Mind2Report, a cognitive deep research agent that emulates the commercial analyst to synthesize expert-level reports. Specifically, it first probes fine-grained intent, then searches web sources and records distilled information on the fly, and subsequently iteratively synthesizes the report. We design Mind2Report as a training-free agentic workflow that augments general large language models (LLMs) with dynamic memory to support these long-form cognitive processes. To rigorously evaluate Mind2Report, we further construct QRC-Eval comprising 200 real-world commercial tasks and establish a holistic evaluation strategy to assess report quality, reliability, and coverage. Experiments demonstrate that Mind2Report outperforms leading baselines, including OpenAI and Gemini deep research agents. Although this is a preliminary study, we expect it to serve as a foundation for advancing the future design of commercial deep research agents. Our code and data are available at https://github.com/Melmaphother/Mind2Report.
Scientific inquiry requires systems-level reasoning that integrates heterogeneous experimental data, cross-domain knowledge, and mechanistic evidence into coherent explanations. While Large Language Models (LLMs) offer inferential capabilities, they often depend on retrieval-augmented contexts that lack structural depth. Traditional Knowledge Graphs (KGs) attempt to bridge this gap, yet their pairwise constraints fail to capture the irreducible higher-order interactions that govern emergent physical behavior. To address this, we introduce a methodology for constructing hypergraph-based knowledge representations that faithfully encode multi-entity relationships. Applied to a corpus of ~1,100 manuscripts on biocomposite scaffolds, our framework constructs a global hypergraph of 161,172 nodes and 320,201 hyperedges, revealing a scale-free topology (power law exponent ~1.23) organized around highly connected conceptual hubs. This representation prevents the combinatorial explosion typical of pairwise expansions and explicitly preserves the co-occurrence context of scientific formulations. We further demonstrate that equipping agentic systems with hypergraph traversal tools, specifically using node-intersection constraints, enables them to bridge semantically distant concepts. By exploiting these higher-order pathways, the system successfully generates grounded mechanistic hypotheses for novel composite materials, such as linking cerium oxide to PCL scaffolds via chitosan intermediates. This work establishes a "teacherless" agentic reasoning system where hypergraph topology acts as a verifiable guardrail, accelerating scientific discovery by uncovering relationships obscured by traditional graph methods.
While multi-agent systems (MAS) have demonstrated superior performance over single-agent approaches in complex reasoning tasks, they often suffer from significant computational inefficiencies. Existing frameworks typically deploy large language models (LLMs) uniformly across all agent roles, failing to account for the varying cognitive demands of different reasoning stages. We address this inefficiency by proposing OI-MAS framework, a novel multi-agent framework that implements an adaptive model-selection policy across a heterogeneous pool of multi-scale LLMs. Specifically, OI-MAS introduces a state-dependent routing mechanism that dynamically selects agent roles and model scales throughout the reasoning process. In addition, we introduce a confidence-aware mechanism that selects appropriate model scales conditioned on task complexity, thus reducing unnecessary reliance on large-scale models. Experimental results show that OI-MAS consistently outperforms baseline multi-agent systems, improving accuracy by up to 12.88\% while reducing cost by up to 79.78\%.
Cross-domain misinformation detection is challenging, as misinformation arises across domains with substantial differences in knowledge and discourse. Existing methods often rely on single-perspective cues and struggle to generalize to challenging or underrepresented domains, while reasoning large language models (LLMs), though effective on complex tasks, are limited to same-distribution data. To address these gaps, we introduce RAAR, the first retrieval-augmented agentic reasoning framework for cross-domain misinformation detection. To enable cross-domain transfer beyond same-distribution assumptions, RAAR retrieves multi-perspective source-domain evidence aligned with each target sample's semantics, sentiment, and writing style. To overcome single-perspective modeling and missing systematic reasoning, RAAR constructs verifiable multi-step reasoning paths through specialized multi-agent collaboration, where perspective-specific agents produce complementary analyses and a summary agent integrates them under verifier guidance. RAAR further applies supervised fine-tuning and reinforcement learning to train a single multi-task verifier to enhance verification and reasoning capabilities. Based on RAAR, we trained the RAAR-8b and RAAR-14b models. Evaluation on three cross-domain misinformation detection tasks shows that RAAR substantially enhances the capabilities of the base models and outperforms other cross-domain methods, advanced LLMs, and LLM-based adaptation approaches. The project will be released at https://github.com/lzw108/RAAR.
As LLM agents transition from digital assistants to physical controllers in autonomous systems and robotics, they face an escalating threat from indirect prompt injection. By embedding adversarial instructions into the results of tool calls, attackers can hijack the agent's decision-making process to execute unauthorized actions. This vulnerability poses a significant risk as agents gain more direct control over physical environments. Existing defense mechanisms against Indirect Prompt Injection (IPI) generally fall into two categories. The first involves training dedicated detection models; however, this approach entails high computational overhead for both training and inference, and requires frequent updates to keep pace with evolving attack vectors. Alternatively, prompt-based methods leverage the inherent capabilities of LLMs to detect or ignore malicious instructions via prompt engineering. Despite their flexibility, most current prompt-based defenses suffer from high Attack Success Rates (ASR), demonstrating limited robustness against sophisticated injection attacks. In this paper, we propose a novel method that provides LLMs with precise data via tool result parsing while effectively filtering out injected malicious code. Our approach achieves competitive Utility under Attack (UA) while maintaining the lowest Attack Success Rate (ASR) to date, significantly outperforming existing methods. Code is available at GitHub.
Recent advances in large language models (LLMs) enable agentic systems trained with reinforcement learning (RL) over multi-turn interaction trajectories, but practical deployment is bottlenecked by rapidly growing textual histories that inflate token budgets and memory usage. We introduce AgentOCR, a framework that exploits the superior information density of visual tokens by representing the accumulated observation-action history as a compact rendered image. To make multi-turn rollouts scalable, AgentOCR proposes segment optical caching. By decomposing history into hashable segments and maintaining a visual cache, this mechanism eliminates redundant re-rendering. Beyond fixed rendering, AgentOCR introduces agentic self-compression, where the agent actively emits a compression rate and is trained with compression-aware reward to adaptively balance task success and token efficiency. We conduct extensive experiments on challenging agentic benchmarks, ALFWorld and search-based QA. Remarkably, results demonstrate that AgentOCR preserves over 95\% of text-based agent performance while substantially reducing token consumption (>50\%), yielding consistent token and memory efficiency. Our further analysis validates a 20x rendering speedup from segment optical caching and the effective strategic balancing of self-compression.
As large language models (LLMs) transition from general knowledge retrieval to complex scientific discovery, their evaluation standards must also incorporate the rigorous norms of scientific inquiry. Existing benchmarks exhibit a critical blind spot: general instruction-following metrics focus on superficial formatting, while domain-specific scientific benchmarks assess only final-answer correctness, often rewarding models that arrive at the right result with the wrong reasons. To address this gap, we introduce scientific instruction following: the capability to solve problems while strictly adhering to the constraints that establish scientific validity. Specifically, we introduce SciIF, a multi-discipline benchmark that evaluates this capability by pairing university-level problems with a fixed catalog of constraints across three pillars: scientific conditions (e.g., boundary checks and assumptions), semantic stability (e.g., unit and symbol conventions), and specific processes(e.g., required numerical methods). Uniquely, SciIF emphasizes auditability, requiring models to provide explicit evidence of constraint satisfaction rather than implicit compliance. By measuring both solution correctness and multi-constraint adherence, SciIF enables finegrained diagnosis of compositional reasoning failures, ensuring that LLMs can function as reliable agents within the strict logical frameworks of science.
LLM agents have emerged as powerful systems for tackling multi-turn tasks by interleaving internal reasoning and external tool interactions. Agentic Reinforcement Learning has recently drawn significant research attention as a critical post-training paradigm to further refine these capabilities. In this paper, we present AT$^2$PO (Agentic Turn-based Policy Optimization via Tree Search), a unified framework for multi-turn agentic RL that addresses three core challenges: limited exploration diversity, sparse credit assignment, and misaligned policy optimization. AT$^2$PO introduces a turn-level tree structure that jointly enables Entropy-Guided Tree Expansion for strategic exploration and Turn-wise Credit Assignment for fine-grained reward propagation from sparse outcomes. Complementing this, we propose Agentic Turn-based Policy Optimization, a turn-level learning objective that aligns policy updates with the natural decision granularity of agentic interactions. ATPO is orthogonal to tree search and can be readily integrated into any multi-turn RL pipeline. Experiments across seven benchmarks demonstrate consistent improvements over the state-of-the-art baseline by up to 1.84 percentage points in average, with ablation studies validating the effectiveness of each component. Our code is available at https://github.com/zzfoutofspace/ATPO.
Multi-agent AI systems have proven effective for complex reasoning. These systems are compounded by specialized agents, which collaborate through explicit communication, but incur substantial computational overhead. A natural question arises: can we achieve similar modularity benefits with a single agent that selects from a library of skills? We explore this question by viewing skills as internalized agent behaviors. From this perspective, a multi-agent system can be compiled into an equivalent single-agent system, trading inter-agent communication for skill selection. Our preliminary experiments suggest this approach can substantially reduce token usage and latency while maintaining competitive accuracy on reasoning benchmarks. However, this efficiency raises a deeper question that has received little attention: how does skill selection scale as libraries grow? Drawing on principles from cognitive science, we propose that LLM skill selection exhibits bounded capacity analogous to human decision-making. We investigate the scaling behavior of skill selection and observe a striking pattern. Rather than degrading gradually, selection accuracy remains stable up to a critical library size, then drops sharply, indicating a phase transition reminiscent of capacity limits in human cognition. Furthermore, we find evidence that semantic confusability among similar skills, rather than library size alone, plays a central role in this degradation. This perspective suggests that hierarchical organization, which has long helped humans manage complex choices, may similarly benefit AI systems. Our initial results with hierarchical routing support this hypothesis. This work opens new questions about the fundamental limits of semantic-based skill selection in LLMs and offers a cognitive-grounded framework and practical guidelines for designing scalable skill-based agents.
Large Language Models (LLMs) suffer from hallucinations and factual inaccuracies, especially in complex reasoning and fact verification tasks. Multi-Agent Debate (MAD) systems aim to improve answer accuracy by enabling multiple LLM agents to engage in dialogue, promoting diverse reasoning and mutual verification. However, existing MAD frameworks primarily rely on internal knowledge or static documents, making them vulnerable to hallucinations. While MADKE introduces external evidence to mitigate this, its one-time retrieval mechanism limits adaptability to new arguments or emerging information during the debate. To address these limitations, We propose Tool-MAD, a multi-agent debate framework that enhances factual verification by assigning each agent a distinct external tool, such as a search API or RAG module. Tool-MAD introduces three key innovations: (1) a multi-agent debate framework where agents leverage heterogeneous external tools, encouraging diverse perspectives, (2) an adaptive query formulation mechanism that iteratively refines evidence retrieval based on the flow of the debate, and (3) the integration of Faithfulness and Answer Relevance scores into the final decision process, allowing the Judge agent to quantitatively assess the coherence and question alignment of each response and effectively detect hallucinations. Experimental results on four fact verification benchmarks demonstrate that Tool-MAD consistently outperforms state-of-the-art MAD frameworks, achieving up to 5.5% accuracy improvement. Furthermore, in medically specialized domains, Tool-MAD exhibits strong robustness and adaptability across various tool configurations and domain conditions, confirming its potential for broader real-world fact-checking applications.
Large language models (LLMs) are increasingly deployed as intelligent agents that reason, plan, and interact with their environments. To effectively scale to long-horizon scenarios, a key capability for such agents is a memory mechanism that can retain, organize, and retrieve past experiences to support downstream decision-making. However, most existing approaches organize and store memories in a flat manner and rely on simple similarity-based retrieval techniques. Even when structured memory is introduced, existing methods often struggle to explicitly capture the logical relationships among experiences or memory units. Moreover, memory access is largely detached from the constructed structure and still depends on shallow semantic retrieval, preventing agents from reasoning logically over long-horizon dependencies. In this work, we propose CompassMem, an event-centric memory framework inspired by Event Segmentation Theory. CompassMem organizes memory as an Event Graph by incrementally segmenting experiences into events and linking them through explicit logical relations. This graph serves as a logic map, enabling agents to perform structured and goal-directed navigation over memory beyond superficial retrieval, progressively gathering valuable memories to support long-horizon reasoning. Experiments on LoCoMo and NarrativeQA demonstrate that CompassMem consistently improves both retrieval and reasoning performance across multiple backbone models.
Despite the rapid proliferation of Role-Playing Agents (RPAs) based on Large Language Models (LLMs), the structural dimensions defining a character's identity remain weakly formalized, often treating characters as arbitrary text inputs. In this paper, we propose the concept of \textbf{Character Identity}, a multidimensional construct that disentangles a character into two distinct layers: \textbf{(1) Parametric Identity}, referring to character-specific knowledge encoded from the LLM's pre-training, and \textbf{(2) Attributive Identity}, capturing fine-grained behavioral properties such as personality traits and moral values. To systematically investigate these layers, we construct a unified character profile schema and generate both Famous and Synthetic characters under identical structural constraints. Our evaluation across single-turn and multi-turn interactions reveals two critical phenomena. First, we identify \textit{"Fame Fades"}: while famous characters hold a significant advantage in initial turns due to parametric knowledge, this edge rapidly vanishes as models prioritize accumulating conversational context over pre-trained priors. Second, we find that \textit{"Nature Remains"}: while models robustly portray general personality traits regardless of polarity, RPA performance is highly sensitive to the valence of morality and interpersonal relationships. Our findings pinpoint negative social natures as the primary bottleneck in RPA fidelity, guiding future character construction and evaluation.
Agentic search has emerged as a promising paradigm for complex information seeking by enabling Large Language Models (LLMs) to interleave reasoning with tool use. However, prevailing systems rely on monolithic agents that suffer from structural bottlenecks, including unconstrained reasoning outputs that inflate trajectories, sparse outcome-level rewards that complicate credit assignment, and stochastic search noise that destabilizes learning. To address these challenges, we propose \textbf{M-ASK} (Multi-Agent Search and Knowledge), a framework that explicitly decouples agentic search into two complementary roles: Search Behavior Agents, which plan and execute search actions, and Knowledge Management Agents, which aggregate, filter, and maintain a compact internal context. This decomposition allows each agent to focus on a well-defined subtask and reduces interference between search and context construction. Furthermore, to enable stable coordination, M-ASK employs turn-level rewards to provide granular supervision for both search decisions and knowledge updates. Experiments on multi-hop QA benchmarks demonstrate that M-ASK outperforms strong baselines, achieving not only superior answer accuracy but also significantly more stable training dynamics.\footnote{The source code for M-ASK is available at https://github.com/chenyiqun/M-ASK.}
Large Language Model-based Multi-Agent Systems (LLM-based MAS), where multiple LLM agents collaborate to solve complex tasks, have shown impressive performance in many areas. However, MAS are typically distributed across different devices or environments, making them vulnerable to perturbations such as agent failures. While existing works have studied the adversarial attacks and corresponding defense strategies, they mainly focus on reactively detecting and mitigating attacks after they occur rather than proactively designing inherently resilient systems. In this work, we study the resilience of LLM-based MAS under perturbations and find that both the communication topology and prompt design significantly influence system resilience. Motivated by these findings, we propose ResMAS: a two-stage framework for enhancing MAS resilience. First, we train a reward model to predict the MAS's resilience, based on which we train a topology generator to automatically design resilient topology for specific tasks through reinforcement learning. Second, we introduce a topology-aware prompt optimization method that refines each agent's prompt based on its connections and interactions with other agents. Extensive experiments across a range of tasks show that our approach substantially improves MAS resilience under various constraints. Moreover, our framework demonstrates strong generalization ability to new tasks and models, highlighting its potential for building resilient MASs.
The proliferation of smart home devices has increased the complexity of controlling and managing them, leading to user fatigue. In this context, large language models (LLMs) offer a promising solution by enabling natural-language interfaces for Internet of Things (IoT) control. However, existing LLM-based approaches suffer from unreliable and inefficient device control due to the non-deterministic nature of LLMs, high inference latency and cost, and limited personalization. To address these challenges, we present IoTGPT, an LLM-based smart home agent designed to execute IoT commands in a reliable, efficient, and personalized manner. Inspired by how humans manage complex tasks, IoTGPT decomposes user instructions into subtasks and memorizes them. By reusing learned subtasks, subsequent instructions can be processed more efficiently with fewer LLM calls, improving reliability and reducing both latency and cost. IoTGPT also supports fine-grained personalization by adapting individual subtasks to user preferences. Our evaluation demonstrates that IoTGPT outperforms baselines in accuracy, latency/cost, and personalization, while reducing user workload.
Large language models (LLMs) achieve strong performance on many tasks, but their progress remains uneven across languages and cultures, often reflecting values latent in English-centric training data. To enable practical cultural alignment, we propose a scalable approach that leverages national social studies curricula as a foundation for culture-aware supervision. We introduce CuCu, an automated multi-agent LLM framework that transforms national textbook curricula into open-ended, culture-specific question-answer pairs. Applying CuCu to the Korean national social studies curriculum, we construct KCaQA, comprising 34.1k open-ended QA pairs. Our quantitative and qualitative analyses suggest that KCaQA covers culture-specific topics and produces responses grounded in local sociocultural contexts.
Large language models (LLMs) offer unprecedented opportunities for analyzing social phenomena at scale. This paper demonstrates the value of LLMs in psychological measurement by (1) compiling the first large-scale dataset of election rumors on a niche alt-tech platform, (2) developing a multistage Rumor Detection Agent that leverages LLMs for high-precision content classification, and (3) quantifying the psychological dynamics of rumor propagation, specifically the "illusory truth effect" in a naturalistic setting. The Rumor Detection Agent combines (i) a synthetic data-augmented, fine-tuned RoBERTa classifier, (ii) precision keyword filtering, and (iii) a two-pass LLM verification pipeline using GPT-4o mini. The findings reveal that sharing probability rises steadily with each additional exposure, providing large-scale empirical evidence for dose-response belief reinforcement in ideologically homogeneous networks. Simulation results further demonstrate rapid contagion effects: nearly one quarter of users become "infected" within just four propagation iterations. Taken together, these results illustrate how LLMs can transform psychological science by enabling the rigorous measurement of belief dynamics and misinformation spread in massive, real-world datasets.
Recent progress in large language model (LLM) agents has largely focused on embedding self-improvement mechanisms inside the agent or searching over many concurrent variants. While these approaches can raise aggregate scores, they often yield unstable and hard-to-audit improvement trajectories, making it difficult to guarantee non-regression or to reason about failures across versions. We reframe agent improvement as \textbf{release engineering}: agents are treated as shippable artifacts, and improvement is externalized into a regression-aware release pipeline. We introduce \textbf{AgentDevel}, a release engineering pipeline that iteratively runs the current agent, produces implementation-blind, symptom-level quality signals from execution traces, synthesizes a single release candidate (RC) via executable diagnosis, and promotes it under flip-centered gating. AgentDevel features three core designs: (i) an implementation-blind LLM critic that characterizes failure appearances without accessing agent internals, (ii) script-based executable diagnosis that aggregates dominant symptom patterns and produces auditable engineering specifications, and (iii) flip-centered gating that prioritizes pass to fail regressions and fail to pass fixes as first-class evidence. Unlike population-based search or in-agent self-refinement, AgentDevel maintains a single canonical version line and emphasizes non-regression as a primary objective. Experiments on execution-heavy benchmarks demonstrate that AgentDevel yields stable improvements with significantly fewer regressions while producing reproducible, auditable artifacts. Overall, AgentDevel provides a practical development discipline for building, debugging, and releasing LLM agents as software development.
While AI innovation accelerates rapidly, the intellectual process behind breakthroughs -- how researchers identify gaps, synthesize prior work, and generate insights -- remains poorly understood. The lack of structured data on scientific reasoning hinders systematic analysis and development of AI research agents. We introduce Sci-Reasoning, the first dataset capturing the intellectual synthesis behind high-quality AI research. Using community-validated quality signals and an LLM-accelerated, human-verified pipeline, we trace Oral and Spotlight papers across NeurIPS, ICML, and ICLR (2023-2025) to its key predecessors, articulating specific reasoning links in a structured format. Our analysis identifies 15 distinct thinking patterns, with three dominant strategies accounting for 52.7%: Gap-Driven Reframing (24.2%), Cross-Domain Synthesis (18.0%), and Representation Shift (10.5%). The most powerful innovation recipes combine multiple patterns: Gap-Driven Reframing + Representation Shift, Cross-Domain Synthesis + Representation Shift, and Gap-Driven Reframing + Cross-Domain Synthesis. This dataset enables quantitative studies of scientific progress and provides structured reasoning trajectories for training the next generation AI research agents.