As Large Language Models (LLMs) gain agentic abilities, they will have to navigate complex multi-agent scenarios, interacting with human users and other agents in cooperative and competitive settings. This will require new reasoning skills, chief amongst them being theory of mind (ToM), or the ability to reason about the "mental" states of other agents. However, ToM and other multi-agent abilities in LLMs are poorly understood, since existing benchmarks suffer from narrow scope, data leakage, saturation, and lack of interactivity. We thus propose Decrypto, a game-based benchmark for multi-agent reasoning and ToM drawing inspiration from cognitive science, computational pragmatics and multi-agent reinforcement learning. It is designed to be as easy as possible in all other dimensions, eliminating confounding factors commonly found in other benchmarks. To our knowledge, it is also the first platform for designing interactive ToM experiments. We validate the benchmark design through comprehensive empirical evaluations of frontier LLMs, robustness studies, and human-AI cross-play experiments. We find that LLM game-playing abilities lag behind humans and simple word-embedding baselines. We then create variants of two classic cognitive science experiments within Decrypto to evaluate three key ToM abilities. Surprisingly, we find that state-of-the-art reasoning models are significantly worse at those tasks than their older counterparts. This demonstrates that Decrypto addresses a crucial gap in current reasoning and ToM evaluations, and paves the path towards better artificial agents.
Large language models (LLMs) excel at reasoning-only tasks, but struggle when reasoning must be tightly coupled with retrieval, as in multi-hop question answering. To overcome these limitations, we introduce a prompting strategy that first decomposes a complex question into smaller steps, then dynamically constructs a database of facts using LLMs, and finally pieces these facts together to solve the question. We show how this three-stage strategy, which we call Memento, can boost the performance of existing prompting strategies across diverse settings. On the 9-step PhantomWiki benchmark, Memento doubles the performance of chain-of-thought (CoT) when all information is provided in context. On the open-domain version of 2WikiMultiHopQA, CoT-RAG with Memento improves over vanilla CoT-RAG by more than 20 F1 percentage points and over the multi-hop RAG baseline, IRCoT, by more than 13 F1 percentage points. On the challenging MuSiQue dataset, Memento improves ReAct by more than 3 F1 percentage points, demonstrating its utility in agentic settings.
Agents based on Large Language Models (LLMs) have demonstrated strong capabilities across a wide range of tasks. However, deploying LLM-based agents in high-stakes domains comes with significant safety and ethical risks. Unethical behavior by these agents can directly result in serious real-world consequences, including physical harm and financial loss. To efficiently steer the ethical behavior of agents, we frame agent behavior steering as a model editing task, which we term Behavior Editing. Model editing is an emerging area of research that enables precise and efficient modifications to LLMs while preserving their overall capabilities. To systematically study and evaluate this approach, we introduce BehaviorBench, a multi-tier benchmark grounded in psychological moral theories. This benchmark supports both the evaluation and editing of agent behaviors across a variety of scenarios, with each tier introducing more complex and ambiguous scenarios. We first demonstrate that Behavior Editing can dynamically steer agents toward the target behavior within specific scenarios. Moreover, Behavior Editing enables not only scenario-specific local adjustments but also more extensive shifts in an agent's global moral alignment. We demonstrate that Behavior Editing can be used to promote ethical and benevolent behavior or, conversely, to induce harmful or malicious behavior. Through comprehensive evaluations on agents based on frontier LLMs, BehaviorBench shows the effectiveness of Behavior Editing across different models and scenarios. Our findings offer key insights into a new paradigm for steering agent behavior, highlighting both the promise and perils of Behavior Editing.
The global demand for sustainable protein sources has accelerated the need for intelligent tools that can rapidly process and synthesise domain-specific scientific knowledge. In this study, we present a proof-of-concept multi-agent Artificial Intelligence (AI) framework designed to support sustainable protein production research, with an initial focus on microbial protein sources. Our Retrieval-Augmented Generation (RAG)-oriented system consists of two GPT-based LLM agents: (1) a literature search agent that retrieves relevant scientific literature on microbial protein production for a specified microbial strain, and (2) an information extraction agent that processes the retrieved content to extract relevant biological and chemical information. Two parallel methodologies, fine-tuning and prompt engineering, were explored for agent optimisation. Both methods demonstrated effectiveness at improving the performance of the information extraction agent in terms of transformer-based cosine similarity scores between obtained and ideal outputs. Mean cosine similarity scores were increased by up to 25%, while universally reaching mean scores of $\geq 0.89$ against ideal output text. Fine-tuning overall improved the mean scores to a greater extent (consistently of $\geq 0.94$) compared to prompt engineering, although lower statistical uncertainties were observed with the latter approach. A user interface was developed and published for enabling the use of the multi-agent AI system, alongside preliminary exploration of additional chemical safety-based search capabilities
Rare diseases collectively affect over 300 million individuals worldwide, yet timely and accurate diagnosis remains a pervasive challenge. This is largely due to their clinical heterogeneity, low individual prevalence, and the limited familiarity most clinicians have with rare conditions. Here, we introduce DeepRare, the first rare disease diagnosis agentic system powered by a large language model (LLM), capable of processing heterogeneous clinical inputs. The system generates ranked diagnostic hypotheses for rare diseases, each accompanied by a transparent chain of reasoning that links intermediate analytic steps to verifiable medical evidence. DeepRare comprises three key components: a central host with a long-term memory module; specialized agent servers responsible for domain-specific analytical tasks integrating over 40 specialized tools and web-scale, up-to-date medical knowledge sources, ensuring access to the most current clinical information. This modular and scalable design enables complex diagnostic reasoning while maintaining traceability and adaptability. We evaluate DeepRare on eight datasets. The system demonstrates exceptional diagnostic performance among 2,919 diseases, achieving 100% accuracy for 1013 diseases. In HPO-based evaluations, DeepRare significantly outperforms other 15 methods, like traditional bioinformatics diagnostic tools, LLMs, and other agentic systems, achieving an average Recall@1 score of 57.18% and surpassing the second-best method (Reasoning LLM) by a substantial margin of 23.79 percentage points. For multi-modal input scenarios, DeepRare achieves 70.60% at Recall@1 compared to Exomiser's 53.20% in 109 cases. Manual verification of reasoning chains by clinical experts achieves 95.40% agreements. Furthermore, the DeepRare system has been implemented as a user-friendly web application http://raredx.cn/doctor.
Ensuring the security of complex system-on-chips (SoCs) designs is a critical imperative, yet traditional verification techniques struggle to keep pace due to significant challenges in automation, scalability, comprehensiveness, and adaptability. The advent of large language models (LLMs), with their remarkable capabilities in natural language understanding, code generation, and advanced reasoning, presents a new paradigm for tackling these issues. Moving beyond monolithic models, an agentic approach allows for the creation of multi-agent systems where specialized LLMs collaborate to solve complex problems more effectively. Recognizing this opportunity, we introduce SV-LLM, a novel multi-agent assistant system designed to automate and enhance SoC security verification. By integrating specialized agents for tasks like verification question answering, security asset identification, threat modeling, test plan and property generation, vulnerability detection, and simulation-based bug validation, SV-LLM streamlines the workflow. To optimize their performance in these diverse tasks, agents leverage different learning paradigms, such as in-context learning, fine-tuning, and retrieval-augmented generation (RAG). The system aims to reduce manual intervention, improve accuracy, and accelerate security analysis, supporting proactive identification and mitigation of risks early in the design cycle. We demonstrate its potential to transform hardware security practices through illustrative case studies and experiments that showcase its applicability and efficacy.
Recent advancements in tool-augmented large language models have enabled them to interact with external tools, enhancing their ability to perform complex user tasks. However, existing approaches overlook the role of personalisation in guiding tool use. This work investigates how user preferences can be effectively integrated into goal-oriented dialogue agents. Through extensive analysis, we identify key weaknesses in the ability of LLMs to personalise tool use. To this end, we introduce \name, a novel solution that enhances personalised tool use by leveraging a structured tagging tool and an uncertainty-based tool detector. TAPS significantly improves the ability of LLMs to incorporate user preferences, achieving the new state-of-the-art for open source models on the NLSI task.
Can we leverage LLMs to model the process of discovering novel language model (LM) architectures? Inspired by real research, we propose a multi-agent LLM approach that simulates the conventional stages of research, from ideation and literature search (proposal stage) to design implementation (code generation), generative pre-training, and downstream evaluation (verification). Using ideas from scaling laws, our system, Genesys, employs a Ladder of Scales approach; new designs are proposed, adversarially reviewed, implemented, and selectively verified at increasingly larger model scales (14M$\sim$350M parameters) with a narrowing budget (the number of models we can train at each scale). To help make discovery efficient and factorizable, Genesys uses a novel genetic programming backbone, which we show has empirical advantages over commonly used direct prompt generation workflows (e.g., $\sim$86\% percentage point improvement in successful design generation, a key bottleneck). We report experiments involving 1,162 newly discovered designs (1,062 fully verified through pre-training) and find the best designs to be highly competitive with known architectures (e.g., outperform GPT2, Mamba2, etc., on 6/9 common benchmarks). We couple these results with comprehensive system-level ablations and formal results, which give broader insights into the design of effective autonomous discovery systems.
We propose PSALM-V, the first autonomous neuro-symbolic learning system able to induce symbolic action semantics (i.e., pre- and post-conditions) in visual environments through interaction. PSALM-V bootstraps reliable symbolic planning without expert action definitions, using LLMs to generate heuristic plans and candidate symbolic semantics. Previous work has explored using large language models to generate action semantics for Planning Domain Definition Language (PDDL)-based symbolic planners. However, these approaches have primarily focused on text-based domains or relied on unrealistic assumptions, such as access to a predefined problem file, full observability, or explicit error messages. By contrast, PSALM-V dynamically infers PDDL problem files and domain action semantics by analyzing execution outcomes and synthesizing possible error explanations. The system iteratively generates and executes plans while maintaining a tree-structured belief over possible action semantics for each action, iteratively refining these beliefs until a goal state is reached. Simulated experiments of task completion in ALFRED demonstrate that PSALM-V increases the plan success rate from 37% (Claude-3.7) to 74% in partially observed setups. Results on two 2D game environments, RTFM and Overcooked-AI, show that PSALM-V improves step efficiency and succeeds in domain induction in multi-agent settings. PSALM-V correctly induces PDDL pre- and post-conditions for real-world robot BlocksWorld tasks, despite low-level manipulation failures from the robot.
Developing effective instruction-following policies in reinforcement learning remains challenging due to the reliance on extensive human-labeled instruction datasets and the difficulty of learning from sparse rewards. In this paper, we propose a novel approach that leverages the capabilities of large language models (LLMs) to automatically generate open-ended instructions retrospectively from previously collected agent trajectories. Our core idea is to employ LLMs to relabel unsuccessful trajectories by identifying meaningful subtasks the agent has implicitly accomplished, thereby enriching the agent's training data and substantially alleviating reliance on human annotations. Through this open-ended instruction relabeling, we efficiently learn a unified instruction-following policy capable of handling diverse tasks within a single policy. We empirically evaluate our proposed method in the challenging Craftax environment, demonstrating clear improvements in sample efficiency, instruction coverage, and overall policy performance compared to state-of-the-art baselines. Our results highlight the effectiveness of utilizing LLM-guided open-ended instruction relabeling to enhance instruction-following reinforcement learning.
Recent advances in Large Language Models (LLMs) have demonstrated strong potential in code generation, yet their effectiveness in quantum computing remains underexplored. This paper benchmarks LLMs for PennyLane-based quantum code generation using real-world challenges from the Quantum Hackathon (QHack). We introduce QHackBench, a novel benchmark dataset derived from QHack competitions, and evaluate model performance under vanilla prompting and Retrieval-Augmented Generation (RAG). Our structured evaluation framework assesses functional correctness, syntactic validity, and execution success across varying challenge difficulties. Results indicate that RAG-enhanced models, supplemented with an augmented PennyLane dataset, approximately generate similar results as the standard prompting, particularly in complex quantum algorithms. Additionally, we introduce a multi-agent evaluation pipeline that iteratively refines incorrect solutions, further enhancing execution success rates. To foster further research, we commit to publicly releasing QHackBench, along with our evaluation framework and experimental results, enabling continued advancements in AI-assisted quantum programming.
We present Prover Agent, a novel AI agent for automated theorem proving that integrates large language models (LLMs) with a formal proof assistant, Lean. Prover Agent coordinates an informal reasoning LLM, a formal prover model, and feedback from Lean while also generating auxiliary lemmas to assist in discovering the overall proof strategy. It achieves an 86.1% success rate on the MiniF2F benchmark, establishing a new state-of-the-art among methods using small language models (SLMs) with a much lower sample budget than previous approaches. We also present case studies illustrating how these generated lemmas contribute to solving challenging problems.
Multi-agent reinforcement learning (MARL) has emerged as a prominent paradigm for increasingly complex tasks. However, joint evolution across heterogeneous agents remains challenging due to cooperative inefficiency and training instability. In this paper, we propose the joint evolution dynamics for MARL called JoyAgents-R1, which first applies Group Relative Policy Optimization (GRPO) to the joint training of heterogeneous multi-agents. By iteratively refining agents' large language models (LLMs) and memories, the method achieves holistic equilibrium with optimal decision-making and memory capabilities. Specifically, JoyAgents-R1 first implements node-wise Monte Carlo sampling on the behavior of each agent across entire reasoning trajectories to enhance GRPO sampling efficiency while maintaining policy diversity. Then, our marginal benefit-driven selection strategy identifies top-$K$ sampling groups with maximal reward fluctuations, enabling targeted agent model updates that improve training stability and maximize joint benefits through cost-effective parameter adjustments. Meanwhile, JoyAgents-R1 introduces an adaptive memory evolution mechanism that repurposes GRPO rewards as cost-free supervisory signals to eliminate repetitive reasoning and accelerate convergence. Experiments across general and domain-specific scenarios demonstrate that JoyAgents-R1 achieves performance comparable to that of larger LLMs while built on smaller open-source models.
Recent advancements in medical Large Language Models (LLMs) have showcased their powerful reasoning and diagnostic capabilities. Despite their success, current unified multimodal medical LLMs face limitations in knowledge update costs, comprehensiveness, and flexibility. To address these challenges, we introduce the Modular Multi-Agent Framework for Multi-Modal Medical Diagnosis (MAM). Inspired by our empirical findings highlighting the benefits of role assignment and diagnostic discernment in LLMs, MAM decomposes the medical diagnostic process into specialized roles: a General Practitioner, Specialist Team, Radiologist, Medical Assistant, and Director, each embodied by an LLM-based agent. This modular and collaborative framework enables efficient knowledge updates and leverages existing medical LLMs and knowledge bases. Extensive experimental evaluations conducted on a wide range of publicly accessible multimodal medical datasets, incorporating text, image, audio, and video modalities, demonstrate that MAM consistently surpasses the performance of modality-specific LLMs. Notably, MAM achieves significant performance improvements ranging from 18% to 365% compared to baseline models. Our code is released at https://github.com/yczhou001/MAM.
This position paper argues that large language model (LLM)-based social simulations should establish clear boundaries to meaningfully contribute to social science research. While LLMs offer promising capabilities for modeling human-like agents compared to traditional agent-based modeling, they face fundamental limitations that constrain their reliability for social pattern discovery. The core issue lies in LLMs' tendency towards an ``average persona'' that lacks sufficient behavioral heterogeneity, a critical requirement for simulating complex social dynamics. We examine three key boundary problems: alignment (simulated behaviors matching real-world patterns), consistency (maintaining coherent agent behavior over time), and robustness (reproducibility under varying conditions). We propose heuristic boundaries for determining when LLM-based simulations can reliably advance social science understanding. We believe that these simulations are more valuable when focusing on (1) collective patterns rather than individual trajectories, (2) agent behaviors aligning with real population averages despite limited variance, and (3) proper validation methods available for testing simulation robustness. We provide a practical checklist to guide researchers in determining the appropriate scope and claims for LLM-based social simulations.
Query rewriting is pivotal for enhancing dense retrieval, yet current methods demand large-scale supervised data or suffer from inefficient reinforcement learning (RL) exploration. In this work, we first establish that guiding Large Language Models (LLMs) with a concise set of expert-crafted strategies, such as semantic expansion and entity disambiguation, substantially improves retrieval effectiveness on challenging benchmarks, including HotpotQA, FEVER, NFCorpus, and SciFact. Building on this insight, we introduce the Strategy-Adaptive Generation Engine (SAGE), which operationalizes these strategies in an RL framework. SAGE introduces two novel reward shaping mechanisms-Strategic Credit Shaping (SCS) and Contrastive Reward Shaping (CRS)-to deliver more informative learning signals. This strategy-guided approach not only achieves new state-of-the-art NDCG@10 results, but also uncovers a compelling emergent behavior: the agent learns to select optimal strategies, reduces unnecessary exploration, and generates concise rewrites, lowering inference cost without sacrificing performance. Our findings demonstrate that strategy-guided RL, enhanced with nuanced reward shaping, offers a scalable, efficient, and more interpretable paradigm for developing the next generation of robust information retrieval systems.
Multi-sensor fusion perception (MSFP) is a key technology for embodied AI, which can serve a variety of downstream tasks (e.g., 3D object detection and semantic segmentation) and application scenarios (e.g., autonomous driving and swarm robotics). Recently, impressive achievements on AI-based MSFP methods have been reviewed in relevant surveys. However, we observe that the existing surveys have some limitations after a rigorous and detailed investigation. For one thing, most surveys are oriented to a single task or research field, such as 3D object detection or autonomous driving. Therefore, researchers in other related tasks often find it difficult to benefit directly. For another, most surveys only introduce MSFP from a single perspective of multi-modal fusion, while lacking consideration of the diversity of MSFP methods, such as multi-view fusion and time-series fusion. To this end, in this paper, we hope to organize MSFP research from a task-agnostic perspective, where methods are reported from various technical views. Specifically, we first introduce the background of MSFP. Next, we review multi-modal and multi-agent fusion methods. A step further, time-series fusion methods are analyzed. In the era of LLM, we also investigate multimodal LLM fusion methods. Finally, we discuss open challenges and future directions for MSFP. We hope this survey can help researchers understand the important progress in MSFP and provide possible insights for future research.
In recent years, Large-Language-Model-driven AI agents have exhibited unprecedented intelligence, flexibility, and adaptability, and are rapidly changing human production and lifestyle. Nowadays, agents are undergoing a new round of evolution. They no longer act as an isolated island like LLMs. Instead, they start to communicate with diverse external entities, such as other agents and tools, to collectively perform more complex tasks. Under this trend, agent communication is regarded as a foundational pillar of the future AI ecosystem, and many organizations intensively begin to design related communication protocols (e.g., Anthropic's MCP and Google's A2A) within the recent few months. However, this new field exposes significant security hazard, which can cause severe damage to real-world scenarios. To help researchers to quickly figure out this promising topic and benefit the future agent communication development, this paper presents a comprehensive survey of agent communication security. More precisely, we first present a clear definition of agent communication and categorize the entire lifecyle of agent communication into three stages: user-agent interaction, agent-agent communication, and agent-environment communication. Next, for each communication phase, we dissect related protocols and analyze its security risks according to the communication characteristics. Then, we summarize and outlook on the possible defense countermeasures for each risk. Finally, we discuss open issues and future directions in this promising research field.
We introduce TAPAS (Task-based Adaptation and Planning using AgentS), a multi-agent framework that integrates Large Language Models (LLMs) with symbolic planning to solve complex tasks without the need for manually defined environment models. TAPAS employs specialized LLM-based agents that collaboratively generate and adapt domain models, initial states, and goal specifications as needed using structured tool-calling mechanisms. Through this tool-based interaction, downstream agents can request modifications from upstream agents, enabling adaptation to novel attributes and constraints without manual domain redefinition. A ReAct (Reason+Act)-style execution agent, coupled with natural language plan translation, bridges the gap between dynamically generated plans and real-world robot capabilities. TAPAS demonstrates strong performance in benchmark planning domains and in the VirtualHome simulated real-world environment.
While Large Language Models (LLMs) possess significant capabilities in open-world agent tasks, they also face challenges in rapidly adapting to new, specialized tasks due to their reliance on static pre-trained knowledge. Traditional methods such as fine-tuning are often costly, data-intensive, and may lead to "catastrophic forgetting." Therefore, we present KnowMap, a novel approach that dynamically constructs a knowledge base from environmental and experiential data. KnowMap fine-tunes a small knowledge-embedding model to equip a larger LLM with valuable task-specific knowledge. Our experiments on the ScienceWorld benchmark demonstrate 17.71% improvement for the performance of gpt-4-turbo model. KnowMap not only provides an efficient and effective means for LLM task-adapting, but also highlights how integrating environmental and experiential knowledge can enhance LLMs' reasoning capabilities.
Accessibility remains a critical concern in today's society, as many technologies are not developed to support the full range of user needs. Existing multi-agent systems (MAS) often cannot provide comprehensive assistance for users in need due to the lack of customization stemming from closed-source designs. Consequently, individuals with disabilities frequently encounter significant barriers when attempting to interact with digital environments. We introduce MATE, a multimodal accessibility MAS, which performs the modality conversions based on the user's needs. The system is useful for assisting people with disabilities by ensuring that data will be converted to an understandable format. For instance, if the user cannot see well and receives an image, the system converts this image to its audio description. MATE can be applied to a wide range of domains, industries, and areas, such as healthcare, and can become a useful assistant for various groups of users. The system supports multiple types of models, ranging from LLM API calling to using custom machine learning (ML) classifiers. This flexibility ensures that the system can be adapted to various needs and is compatible with a wide variety of hardware. Since the system is expected to run locally, it ensures the privacy and security of sensitive information. In addition, the framework can be effectively integrated with institutional technologies (e.g., digital healthcare service) for real-time user assistance. Furthermore, we introduce ModCon-Task-Identifier, a model that is capable of extracting the precise modality conversion task from the user input. Numerous experiments show that ModCon-Task-Identifier consistently outperforms other LLMs and statistical models on our custom data. Our code and data are publicly available at https://github.com/AlgazinovAleksandr/Multi-Agent-MATE.
LLMs' reliance on static knowledge and fragile tool invocation severely hinders the orchestration of complex, heterogeneous toolchains, particularly at large scales. Existing methods typically use rigid single-path execution, resulting in poor error recovery and exponentially growing search spaces. We introduce NaviAgent, a graph-navigated bilevel planning architecture for robust function calling, comprising a Multi-Path Decider and Graph-Encoded Navigator. As an LLM-powered agent, the Multi-Path Decider defines a four-dimensional decision space and continuously perceives environmental states, dynamically selecting the optimal action to fully cover all tool invocation scenarios. The Graph-Encoded Navigator constructs a Tool Dependency Heterogeneous Graph (TDHG), where node embeddings explicitly fuse API schema structure with historical invocation behavior. It also integrates a novel heuristic search strategy that guides the Decider toward efficient and highly successful toolchains, even for unseen tool combinations. Experiments show that NaviAgent consistently achieves the highest task success rate (TSR) across all foundation models and task complexities, outperforming the average baselines (ReAct, ToolLLM, {\alpha}-UMI) by 13.5%, 16.4%, and 19.0% on Qwen2.5-14B, Qwen2.5-32B, and Deepseek-V3, respectively. Its execution steps are typically within one step of the most efficient baseline, ensuring a strong balance between quality and efficiency. Notably, a fine-tuned Qwen2.5-14B model achieves a TSR of 49.5%, surpassing the much larger 32B model (44.9%) under our architecture. Incorporating the Graph-Encoded Navigator further boosts TSR by an average of 2.4 points, with gains up over 9 points on complex tasks for larger models (Deepseek-V3 and GPT-4o), highlighting its essential role in toolchain orchestration.
Large Language Models (LLMs) are rapidly transforming education by enabling rich conversational learning experiences. This article provides a comprehensive review of how LLM-based conversational agents are being used in higher education, with extensions to secondary and lifelong learning contexts. We synthesize existing literature on LLMs in education and theories of conversational and dialogic pedagogy - including Vygotsky's sociocultural learning (scaffolding and the Zone of Proximal Development), the Socratic method, and Laurillard's conversational framework - and examine how prompting strategies and retrieval-augmented generation (RAG) can align LLM behaviors with these pedagogical theories, and how it can support personalized, adaptive learning. We map educational theories to LLM capabilities, highlighting where LLM-driven dialogue supports established learning principles and where it challenges or falls short of traditional pedagogical assumptions. Notable gaps in applying prior theories to LLMs are identified, such as the models tendency to provide direct answers instead of fostering co-construction of knowledge, and the need to account for the constant availability and broad but non-human expertise of LLM tutors. In response, we propose practical strategies to better align LLM interactions with sound pedagogy - for example, designing prompts that encourage Socratic questioning, scaffolded guidance, and student reflection, as well as integrating retrieval mechanisms to ensure accuracy and contextual relevance. Our aim is to bridge the gap between educational theory and the emerging practice of AI-driven conversational learning, offering insights and tools for making LLM-based dialogues more educationally productive and theory-aligned.
Refactoring is a constant activity in software development and maintenance. Scale and maintain software systems are based on code refactoring. However, this process is still labor intensive, as it requires programmers to analyze the codebases in detail to avoid introducing new defects. In this research, we put forward a large language model (LLM)-based multi-agent system to automate the refactoring process on Haskell code. The objective of this research is to evaluate the effect of LLM-based agents in performing structured and semantically accurate refactoring on Haskell code. Our proposed multi-agent system based on specialized agents with distinct roles, including code analysis, refactoring execution, verification, and debugging. To test the effectiveness and practical applicability of the multi-agent system, we conducted evaluations using different open-source Haskell codebases. The results of the experiments carried out showed that the proposed LLM-based multi-agent system could average 11.03% decreased complexity in code, an improvement of 22.46% in overall code quality, and increase performance efficiency by an average of 13.27%. Furthermore, memory allocation was optimized by up to 14.57%. These results highlight the ability of LLM-based multi-agent in managing refactoring tasks targeted toward functional programming paradigms. Our findings hint that LLM-based multi-agent systems integration into the refactoring of functional programming languages can enhance maintainability and support automated development workflows.
Vision-and-Language Navigation (VLN) in large-scale urban environments requires embodied agents to ground linguistic instructions in complex scenes and recall relevant experiences over extended time horizons. Prior modular pipelines offer interpretability but lack unified memory, while end-to-end (M)LLM agents excel at fusing vision and language yet remain constrained by fixed context windows and implicit spatial reasoning. We introduce \textbf{Mem4Nav}, a hierarchical spatial-cognition long-short memory system that can augment any VLN backbone. Mem4Nav fuses a sparse octree for fine-grained voxel indexing with a semantic topology graph for high-level landmark connectivity, storing both in trainable memory tokens embedded via a reversible Transformer. Long-term memory (LTM) compresses and retains historical observations at both octree and graph nodes, while short-term memory (STM) caches recent multimodal entries in relative coordinates for real-time obstacle avoidance and local planning. At each step, STM retrieval sharply prunes dynamic context, and, when deeper history is needed, LTM tokens are decoded losslessly to reconstruct past embeddings. Evaluated on Touchdown and Map2Seq across three backbones (modular, state-of-the-art VLN with prompt-based LLM, and state-of-the-art VLN with strided-attention MLLM), Mem4Nav yields 7-13 pp gains in Task Completion, sufficient SPD reduction, and >10 pp nDTW improvement. Ablations confirm the indispensability of both the hierarchical map and dual memory modules. Our codes are open-sourced via https://github.com/tsinghua-fib-lab/Mem4Nav.
Multimodal sarcasm understanding is a high-order cognitive task. Although large language models (LLMs) have shown impressive performance on many downstream NLP tasks, growing evidence suggests that they struggle with sarcasm understanding. In this paper, we propose Commander-GPT, a modular decision routing framework inspired by military command theory. Rather than relying on a single LLM's capability, Commander-GPT orchestrates a team of specialized LLM agents where each agent will be selectively assigned to a focused sub-task such as context modeling, sentiment analysis, etc. Their outputs are then routed back to the commander, which integrates the information and performs the final sarcasm judgment. To coordinate these agents, we introduce three types of centralized commanders: (1) a trained lightweight encoder-based commander (e.g., multi-modal BERT); (2) four small autoregressive language models, serving as moderately capable commanders (e.g., DeepSeek-VL); (3) two large LLM-based commander (Gemini Pro and GPT-4o) that performs task routing, output aggregation, and sarcasm decision-making in a zero-shot fashion. We evaluate Commander-GPT on the MMSD and MMSD 2.0 benchmarks, comparing five prompting strategies. Experimental results show that our framework achieves 4.4% and 11.7% improvement in F1 score over state-of-the-art (SoTA) baselines on average, demonstrating its effectiveness.
Software engineering (SWE) has recently emerged as a crucial testbed for next-generation LLM agents, demanding inherent capabilities in two critical dimensions: sustained iterative problem-solving (e.g., >50 interaction rounds) and long-context dependency resolution (e.g., >32k tokens). However, the data curation process in SWE remains notoriously time-consuming, as it heavily relies on manual annotation for code file filtering and the setup of dedicated runtime environments to execute and validate unit tests. Consequently, most existing datasets are limited to only a few thousand GitHub-sourced instances. To this end, we propose an incremental, automated data-curation pipeline that systematically scales both the volume and diversity of SWE datasets. Our dataset comprises 10,169 real-world Python task instances from 2,531 distinct GitHub repositories, each accompanied by a task specified in natural language and a dedicated runtime-environment image for automated unit-test validation. We have carefully curated over 8,000 successfully runtime-validated training trajectories from our proposed SWE dataset. When fine-tuning the Skywork-SWE model on these trajectories, we uncover a striking data scaling phenomenon: the trained model's performance for software engineering capabilities in LLMs continues to improve as the data size increases, showing no signs of saturation. Notably, our Skywork-SWE model achieves 38.0% pass@1 accuracy on the SWE-bench Verified benchmark without using verifiers or multiple rollouts, establishing a new state-of-the-art (SOTA) among the Qwen2.5-Coder-32B-based LLMs built on the OpenHands agent framework. Furthermore, with the incorporation of test-time scaling techniques, the performance further improves to 47.0% accuracy, surpassing the previous SOTA results for sub-32B parameter models. We release the Skywork-SWE-32B model checkpoint to accelerate future research.
Multi-agent techniques such as role playing or multi-turn debates have been shown to be effective in improving the performance of large language models (LLMs) in downstream tasks. Despite their differences in workflows, existing LLM-based multi-agent systems mostly use natural language for agent communication. While this is appealing for its simplicity and interpretability, it also introduces inevitable information loss as one model must down sample its continuous state vectors to concrete tokens before transferring them to the other model. Such losses are particularly significant when the information to transfer is not simple facts, but reasoning logics or abstractive thoughts. To tackle this problem, we propose a new communication protocol that transfers both natural language tokens and token-wise state transition trajectory from one agent to another. Particularly, compared to the actual state value, we find that the sequence of state changes in LLMs after generating each token can better reflect the information hidden behind the inference process, so we propose a State Delta Encoding (SDE) method to represent state transition trajectories. The experimental results show that multi-agent systems with SDE achieve SOTA performance compared to other communication protocols, particularly in tasks that involve complex reasoning. This shows the potential of communication augmentation for LLM-based multi-agent systems.
Large language models (LLMs) often struggle with mathematical problems that require exact computation or multi-step algebraic reasoning. Tool-integrated reasoning (TIR) offers a promising solution by leveraging external tools such as code interpreters to ensure correctness, but it introduces inference-time dependencies that hinder scalability and deployment. In this work, we propose a new paradigm for distilling tool knowledge into LLMs purely through natural language. We first construct a Solver Agent that solves math problems by interleaving planning, symbolic tool calls, and reflective reasoning. Then, using a back-translation pipeline powered by multiple LLM-based agents, we convert interleaved TIR traces into natural language reasoning traces. A Translator Agent generates explanations for individual tool calls, while a Rephrase Agent merges them into a fluent and globally coherent narrative. Empirically, we show that fine-tuning a small open-source model on these synthesized traces enables it to internalize both tool knowledge and structured reasoning patterns, yielding gains on competition-level math benchmarks without requiring tool access at inference.
Traditional control system design, reliant on expert knowledge and precise models, struggles with complex, nonlinear, or uncertain dynamics. This paper introduces AgenticControl, a novel multi-agent framework that automates controller design using coordinated Large Language Model (LLM) agents. Through structured JSON communication, these agents handle tasks including controller selection, scenario design, parameter optimization, performance evaluation, and decision-making. Through an actor-critic optimization approach, the system iteratively improves performance while progressing through scenarios of increasing complexity to ensure robustness under nominal conditions, measurement noise, actuator disturbances, and parametric uncertainties. Key innovations include structured multi-agent collaboration, robust optimization mechanisms, and real-time adaptability via in-context learning. Validated across four diverse control systems, namely, DC Motor Position control, Ball and Beam, Inverted Pendulum, and Double Inverted Pendulum, the framework achieves competitive performance against classical methods. Its Full State Feedback solution closely matches Linear Quadratic Regulator (LQR) results, while the designed PID controller significantly outperforming MATLAB's PIDTuner, reducing PID tracking error by 55% through adaptive parameter exploration. A comparative study of five LLM models reveals distinct optimization profiles, with DeepSeek achieving the fastest convergence. This work demonstrates the potential of LLM-driven control design, paving the way for advanced techniques like model predictive control and reinforcement learning.