We present DELIVER (Directed Execution of Language-instructed Item Via Engineered Relay), a fully integrated framework for cooperative multi-robot pickup and delivery driven by natural language commands. DELIVER unifies natural language understanding, spatial decomposition, relay planning, and motion execution to enable scalable, collision-free coordination in real-world settings. Given a spoken or written instruction, a lightweight instance of LLaMA3 interprets the command to extract pickup and delivery locations. The environment is partitioned using a Voronoi tessellation to define robot-specific operating regions. Robots then compute optimal relay points along shared boundaries and coordinate handoffs. A finite-state machine governs each robot's behavior, enabling robust execution. We implement DELIVER on the MultiTRAIL simulation platform and validate it in both ROS2-based Gazebo simulations and real-world hardware using TurtleBot3 robots. Empirical results show that DELIVER maintains consistent mission cost across varying team sizes while reducing per-agent workload by up to 55% compared to a single-agent system. Moreover, the number of active relay agents remains low even as team size increases, demonstrating the system's scalability and efficient agent utilization. These findings underscore DELIVER's modular and extensible architecture for language-guided multi-robot coordination, advancing the frontiers of cyber-physical system integration.
Large language model (LLM)-based agents have demonstrated remarkable capabilities in decision-making tasks, but struggle significantly with complex, long-horizon planning scenarios. This arises from their lack of macroscopic guidance, causing disorientation and failures in complex tasks, as well as insufficient continuous oversight during execution, rendering them unresponsive to environmental changes and prone to deviations. To tackle these challenges, we introduce HiPlan, a hierarchical planning framework that provides adaptive global-local guidance to boost LLM-based agents'decision-making. HiPlan decomposes complex tasks into milestone action guides for general direction and step-wise hints for detailed actions. During the offline phase, we construct a milestone library from expert demonstrations, enabling structured experience reuse by retrieving semantically similar tasks and milestones. In the execution phase, trajectory segments from past milestones are dynamically adapted to generate step-wise hints that align current observations with the milestone objectives, bridging gaps and correcting deviations. Extensive experiments across two challenging benchmarks demonstrate that HiPlan substantially outperforms strong baselines, and ablation studies validate the complementary benefits of its hierarchical components.
Structured, procedural reasoning is essential for Large Language Models (LLMs), especially in mathematics. While post-training methods have improved LLM performance, they still fall short in capturing deep procedural logic on complex tasks. To tackle the issue, in this paper, we first investigate this limitation and uncover a novel finding: a Scaling Law by Difficulty, which reveals that model performance follows a U-shaped curve with respect to training data complexity -- excessive low-difficulty data impedes abstraction, while high-difficulty data significantly enhances reasoning ability. Motivated by this, we propose the Structured Solution Template (SST) framework, which uses solution templates and a curriculum of varied difficulty to explicitly teach procedural reasoning. Specifically, SST comprises (1) fine-tuning with structured solution-template chains and dynamically weighted loss to prioritize procedural logic, (2) prompt-time injection of solution templates as cognitive scaffolds to guide inference, and (3) integrated curriculum fine-tuning that explicitly teaches the model to self-plan - execute - self-correct. Experiments on GSM8K, AIME24, and new Dynamic En benchmark show that SST significantly improves both accuracy and efficiency, especially on harder problems.
Existing tasks fall short in evaluating reasoning ability of Large Language Models (LLMs) in an interactive, unknown environment. This deficiency leads to the isolated assessment of deductive, inductive, and abductive reasoning, neglecting the integrated reasoning process that is indispensable for humans discovery of real world. We introduce a novel evaluation paradigm, \textit{black-box interaction}, to tackle this challenge. A black-box is defined by a hidden function that maps a specific set of inputs to outputs. LLMs are required to unravel the hidden function behind the black-box by interacting with it in given exploration turns, and reasoning over observed input-output pairs. Leveraging this idea, we build the \textsc{Oracle} benchmark which comprises 6 types of black-box task and 96 black-boxes. 19 modern LLMs are benchmarked. o3 ranks first in 5 of the 6 tasks, achieving over 70\% accuracy on most easy black-boxes. But it still struggles with some hard black-box tasks, where its average performance drops below 40\%. Further analysis indicates a universal difficulty among LLMs: They lack the high-level planning capability to develop efficient and adaptive exploration strategies for hypothesis refinement.
Minecraft, as an open-world virtual interactive environment, has become a prominent platform for research on agent decision-making and execution. Existing works primarily adopt a single Large Language Model (LLM) agent to complete various in-game tasks. However, for complex tasks requiring lengthy sequences of actions, single-agent approaches often face challenges related to inefficiency and limited fault tolerance. Despite these issues, research on multi-agent collaboration remains scarce. In this paper, we propose CausalMACE, a holistic causality planning framework designed to enhance multi-agent systems, in which we incorporate causality to manage dependencies among subtasks. Technically, our proposed framework introduces two modules: an overarching task graph for global task planning and a causality-based module for dependency management, where inherent rules are adopted to perform causal intervention. Experimental results demonstrate our approach achieves state-of-the-art performance in multi-agent cooperative tasks of Minecraft.
Efficient querying and analysis of large tabular datasets remain significant challenges, especially for users without expertise in programming languages like SQL. Text-to-SQL approaches have shown promising performance on benchmark data; however, they inherit SQL's drawbacks, including inefficiency with large datasets and limited support for complex data analyses beyond basic querying. We propose a novel framework that transforms natural language queries into query plans. Our solution is implemented outside traditional databases, allowing us to support classical SQL commands while avoiding SQL's inherent limitations. Additionally, we enable complex analytical functions, such as principal component analysis and anomaly detection, providing greater flexibility and extensibility than traditional SQL capabilities. We leverage LLMs to iteratively interpret queries and construct operation sequences, addressing computational complexity by incrementally building solutions. By executing operations directly on the data, we overcome context length limitations without requiring the entire dataset to be processed by the model. We validate our framework through experiments on both standard databases and large scientific tables, demonstrating its effectiveness in handling extensive datasets and performing sophisticated data analyses.
LLM app stores are quickly emerging as platforms that gather a wide range of intelligent applications based on LLMs, giving users many choices for content creation, coding support, education, and more. However, the current methods for ranking and recommending apps in these stores mostly rely on static metrics like user activity and favorites, which makes it hard for users to efficiently find high-quality apps. To address these challenges, we propose LaQual, an automated framework for evaluating the quality of LLM apps. LaQual consists of three main stages: first, it labels and classifies LLM apps in a hierarchical way to accurately match them to different scenarios; second, it uses static indicators, such as time-weighted user engagement and functional capability metrics, to filter out low-quality apps; and third, it conducts a dynamic, scenario-adaptive evaluation, where the LLM itself generates scenario-specific evaluation metrics, scoring rules, and tasks for a thorough quality assessment. Experiments on a popular LLM app store show that LaQual is effective. Its automated scores are highly consistent with human judgments (with Spearman's rho of 0.62 and p=0.006 in legal consulting, and rho of 0.60 and p=0.009 in travel planning). By effectively screening, LaQual can reduce the pool of candidate LLM apps by 66.7% to 81.3%. User studies further confirm that LaQual significantly outperforms baseline systems in decision confidence, comparison efficiency (with average scores of 5.45 compared to 3.30), and the perceived value of its evaluation reports (4.75 versus 2.25). Overall, these results demonstrate that LaQual offers a scalable, objective, and user-centered solution for finding and recommending high-quality LLM apps in real-world use cases.
Large language model (LLM)-powered agents are increasingly used to plan and execute scientific workflows, yet most research cyberinfrastructure (CI) exposes heterogeneous APIs and implements security models that present barriers for use by agents. We report on our experience using the Model Context Protocol (MCP) as a unifying interface that makes research capabilities discoverable, invokable, and composable. Our approach is pragmatic: we implement thin MCP servers over mature services, including Globus Transfer, Compute, and Search; status APIs exposed by computing facilities; Octopus event fabric; and domain-specific tools such as Garden and Galaxy. We use case studies in computational chemistry, bioinformatics, quantum chemistry, and filesystem monitoring to illustrate how this MCP-oriented architecture can be used in practice. We distill lessons learned and outline open challenges in evaluation and trust for agent-led science.
Modern large language models (LLMs) have demonstrated impressive performance across a wide range of multi-step reasoning tasks. Recent work suggests that LLMs may perform planning - selecting a future target token in advance and generating intermediate tokens that lead towards it - rather than merely improvising one token at a time. However, existing studies assume fixed planning horizons and often focus on single prompts or narrow domains. To distinguish planning from improvisation across models and tasks, we present formal and causally grounded criteria for detecting planning and operationalize them as a semi-automated annotation pipeline. We apply this pipeline to both base and instruction-tuned Gemma-2-2B models on the MBPP code generation benchmark and a poem generation task where Claude 3.5 Haiku was previously shown to plan. Our findings show that planning is not universal: unlike Haiku, Gemma-2-2B solves the same poem generation task through improvisation, and on MBPP it switches between planning and improvisation across similar tasks and even successive token predictions. We further show that instruction tuning refines existing planning behaviors in the base model rather than creating them from scratch. Together, these studies provide a reproducible and scalable foundation for mechanistic studies of planning in LLMs.
ACL configuration is essential for managing network flow reachability, yet its complexity grows significantly with topologies and pre-existing rules. To carry out ACL configuration, the operator needs to (1) understand the new configuration policies or intents and translate them into concrete ACL rules, (2) check and resolve any conflicts between the new and existing rules, and (3) deploy them across the network. Existing systems rely heavily on manual efforts for these tasks, especially for the first two, which are tedious, error-prone, and impractical to scale. We propose Xumi to tackle this problem. Leveraging LLMs with domain knowledge of the target network, Xumi automatically and accurately translates the natural language intents into complete ACL rules to reduce operators' manual efforts. Xumi then detects all potential conflicts between new and existing rules and generates resolved intents for deployment with operators' guidance, and finally identifies the best deployment plan that minimizes the rule additions while satisfying all intents. Evaluation shows that Xumi accelerates the entire configuration pipeline by over 10x compared to current practices, addresses O(100) conflicting ACLs and reduces rule additions by ~40% in modern cloud network.
The development and application of large language models (LLM) have demonstrated that foundational models can be utilized to solve a wide array of tasks. However, their performance in multi-agent path finding (MAPF) tasks has been less than satisfactory, with only a few studies exploring this area. MAPF is a complex problem requiring both planning and multi-agent coordination. To improve the performance of LLM in MAPF tasks, we propose a novel framework, LLM-NAR, which leverages neural algorithmic reasoners (NAR) to inform LLM for MAPF. LLM-NAR consists of three key components: an LLM for MAPF, a pre-trained graph neural network-based NAR, and a cross-attention mechanism. This is the first work to propose using a neural algorithmic reasoner to integrate GNNs with the map information for MAPF, thereby guiding LLM to achieve superior performance. LLM-NAR can be easily adapted to various LLM models. Both simulation and real-world experiments demonstrate that our method significantly outperforms existing LLM-based approaches in solving MAPF problems.
Accurately predicting travel mode choice is essential for effective transportation planning, yet traditional statistical and machine learning models are constrained by rigid assumptions, limited contextual reasoning, and reduced generalizability. This study explores the potential of Large Language Models (LLMs) as a more flexible and context-aware approach to travel mode choice prediction, enhanced by Retrieval-Augmented Generation (RAG) to ground predictions in empirical data. We develop a modular framework for integrating RAG into LLM-based travel mode choice prediction and evaluate four retrieval strategies: basic RAG, RAG with balanced retrieval, RAG with a cross-encoder for re-ranking, and RAG with balanced retrieval and cross-encoder for re-ranking. These strategies are tested across three LLM architectures (OpenAI GPT-4o, o4-mini, and o3) to examine the interaction between model reasoning capabilities and retrieval methods. Using the 2023 Puget Sound Regional Household Travel Survey data, we conduct a series of experiments to evaluate model performance. The results demonstrate that RAG substantially enhances predictive accuracy across a range of models. Notably, the GPT-4o model combined with balanced retrieval and cross-encoder re-ranking achieves the highest accuracy of 80.8%, exceeding that of conventional statistical and machine learning baselines. Furthermore, LLM-based models exhibit superior generalization abilities relative to these baselines. Findings highlight the critical interplay between LLM reasoning capabilities and retrieval strategies, demonstrating the importance of aligning retrieval strategies with model capabilities to maximize the potential of LLM-based travel behavior modeling.
Despite the remarkable capabilities of text-to-image (T2I) generation models, real-world applications often demand fine-grained, iterative image editing that existing methods struggle to provide. Key challenges include granular instruction understanding, robust context preservation during modifications, and the lack of intelligent feedback mechanisms for iterative refinement. This paper introduces RefineEdit-Agent, a novel, training-free intelligent agent framework designed to address these limitations by enabling complex, iterative, and context-aware image editing. RefineEdit-Agent leverages the powerful planning capabilities of Large Language Models (LLMs) and the advanced visual understanding and evaluation prowess of Vision-Language Large Models (LVLMs) within a closed-loop system. Our framework comprises an LVLM-driven instruction parser and scene understanding module, a multi-level LLM-driven editing planner for goal decomposition, tool selection, and sequence generation, an iterative image editing module, and a crucial LVLM-driven feedback and evaluation loop. To rigorously evaluate RefineEdit-Agent, we propose LongBench-T2I-Edit, a new benchmark featuring 500 initial images with complex, multi-turn editing instructions across nine visual dimensions. Extensive experiments demonstrate that RefineEdit-Agent significantly outperforms state-of-the-art baselines, achieving an average score of 3.67 on LongBench-T2I-Edit, compared to 2.29 for Direct Re-Prompting, 2.91 for InstructPix2Pix, 3.16 for GLIGEN-based Edit, and 3.39 for ControlNet-XL. Ablation studies, human evaluations, and analyses of iterative refinement, backbone choices, tool usage, and robustness to instruction complexity further validate the efficacy of our agentic design in delivering superior edit fidelity and context preservation.
LLM agents are increasingly deployed to plan, retrieve, and write with tools, yet evaluation still leans on static benchmarks and small human studies. We present the Agent-Testing Agent (ATA), a meta-agent that combines static code analysis, designer interrogation, literature mining, and persona-driven adversarial test generation whose difficulty adapts via judge feedback. Each dialogue is scored with an LLM-as-a-Judge (LAAJ) rubric and used to steer subsequent tests toward the agent's weakest capabilities. On a travel planner and a Wikipedia writer, the ATA surfaces more diverse and severe failures than expert annotators while matching severity, and finishes in 20--30 minutes versus ten-annotator rounds that took days. Ablating code analysis and web search increases variance and miscalibration, underscoring the value of evidence-grounded test generation. The ATA outputs quantitative metrics and qualitative bug reports for developers. We release the full methodology and open-source implementation for reproducible agent testing: https://github.com/KhalilMrini/Agent-Testing-Agent
Mock trial has long served as an important platform for legal professional training and education. It not only helps students learn about realistic trial procedures, but also provides practical value for case analysis and judgment prediction. Traditional mock trials are difficult to access by the public because they rely on professional tutors and human participants. Fortunately, the rise of large language models (LLMs) provides new opportunities for creating more accessible and scalable court simulations. While promising, existing research mainly focuses on agent construction while ignoring the systematic design and evaluation of court simulations, which are actually more important for the credibility and usage of court simulation in practice. To this end, we present the first court simulation framework -- SimCourt -- based on the real-world procedure structure of Chinese courts. Our framework replicates all 5 core stages of a Chinese trial and incorporates 5 courtroom roles, faithfully following the procedural definitions in China. To simulate trial participants with different roles, we propose and craft legal agents equipped with memory, planning, and reflection abilities. Experiment on legal judgment prediction show that our framework can generate simulated trials that better guide the system to predict the imprisonment, probation, and fine of each case. Further annotations by human experts show that agents' responses under our simulation framework even outperformed judges and lawyers from the real trials in many scenarios. These further demonstrate the potential of LLM-based court simulation.
The pursuit of human-level artificial intelligence (AI) has significantly advanced the development of autonomous agents and Large Language Models (LLMs). LLMs are now widely utilized as decision-making agents for their ability to interpret instructions, manage sequential tasks, and adapt through feedback. This review examines recent developments in employing LLMs as autonomous agents and tool users and comprises seven research questions. We only used the papers published between 2023 and 2025 in conferences of the A* and A rank and Q1 journals. A structured analysis of the LLM agents' architectural design principles, dividing their applications into single-agent and multi-agent systems, and strategies for integrating external tools is presented. In addition, the cognitive mechanisms of LLM, including reasoning, planning, and memory, and the impact of prompting methods and fine-tuning procedures on agent performance are also investigated. Furthermore, we evaluated current benchmarks and assessment protocols and have provided an analysis of 68 publicly available datasets to assess the performance of LLM-based agents in various tasks. In conducting this review, we have identified critical findings on verifiable reasoning of LLMs, the capacity for self-improvement, and the personalization of LLM-based agents. Finally, we have discussed ten future research directions to overcome these gaps.
Large Language Model (LLM)-enabled agents are rapidly emerging across a wide range of applications, but their deployment introduces vulnerabilities with security implications. While prior work has examined prompt-based attacks (e.g., prompt injection) and data-oriented threats (e.g., data exfiltration), time-of-check to time-of-use (TOCTOU) remain largely unexplored in this context. TOCTOU arises when an agent validates external state (e.g., a file or API response) that is later modified before use, enabling practical attacks such as malicious configuration swaps or payload injection. In this work, we present the first study of TOCTOU vulnerabilities in LLM-enabled agents. We introduce TOCTOU-Bench, a benchmark with 66 realistic user tasks designed to evaluate this class of vulnerabilities. As countermeasures, we adapt detection and mitigation techniques from systems security to this setting and propose prompt rewriting, state integrity monitoring, and tool-fusing. Our study highlights challenges unique to agentic workflows, where we achieve up to 25% detection accuracy using automated detection methods, a 3% decrease in vulnerable plan generation, and a 95% reduction in the attack window. When combining all three approaches, we reduce the TOCTOU vulnerabilities from an executed trajectory from 12% to 8%. Our findings open a new research direction at the intersection of AI safety and systems security.
Due to the rapid pace of electrification and decarbonization, distribution grid (DG) operation and planning are becoming more complex, necessitating advanced computational analyses to ensure grid reliability and resilience. State-of-the-art DG analyses rely on disparate workflows of complex models, functions, and data pipelines, which require expert knowledge and are challenging to automate. Many small-scale utilities and cooperatives lack a large R&D workforce and therefore cannot use advanced analysis at scale. To address this gap, we develop a novel agentic AI system, PowerChain, to solve unseen DG analysis tasks via automated agentic orchestration and large language models (LLMs) function-calling. Given a natural language query, PowerChain dynamically generates and executes an ordered sequence of domain-aware functions guided by the semantics of an expert-built power systems function pool and a select reference set of known, expert-generated workflow-query pairs. Our results show that PowerChain can produce expert-level workflows with both GPT-5 and open-source Qwen models on complex, unseen DG analysis tasks operating on real utility data.
Recent advances in generalist multi-agent systems (MAS) have largely followed a context-engineering plus centralized paradigm, where a planner agent coordinates multiple worker agents through unidirectional prompt passing. While effective under strong planner models, this design suffers from two critical limitations: (1) strong dependency on the planner's capability, which leads to degraded performance when a smaller LLM powers the planner; and (2) limited inter-agent communication, where collaboration relies on costly prompt concatenation and context injection, introducing redundancy and information loss. To address these challenges, we propose Anemoi, a semi-centralized MAS built on the Agent-to-Agent (A2A) communication MCP server from Coral Protocol. Unlike traditional designs, Anemoi enables structured and direct inter-agent collaboration, allowing all agents to monitor progress, assess results, identify bottlenecks, and propose refinements in real time. This paradigm reduces reliance on a single planner, supports adaptive plan updates, and minimizes redundant context passing, resulting in more scalable and cost-efficient execution. Evaluated on the GAIA benchmark, Anemoi achieved 52.73\% accuracy with a small LLM (GPT-4.1-mini) as the planner, surpassing the strongest open-source baseline OWL (43.63\%) by +9.09\% under identical LLM settings. Our implementation is publicly available at https://github.com/Coral-Protocol/Anemoi.
Table understanding is key to addressing challenging downstream tasks such as table-based question answering and fact verification. Recent works have focused on leveraging Chain-of-Thought and question decomposition to solve complex questions requiring multiple operations on tables. However, these methods often suffer from a lack of explicit long-term planning and weak inter-step connections, leading to miss constraints within questions. In this paper, we propose leveraging the long-term planning capabilities of large language models (LLMs) to enhance table understanding. Our approach enables the execution of a long-term plan, where the steps are tightly interconnected and serve the ultimate goal, an aspect that methods based on Chain-of-Thought and question decomposition lack. In addition, our method effectively minimizes the inclusion of unnecessary details in the process of solving the next short-term goals, a limitation of methods based on Chain-of-Thought. Extensive experiments demonstrate that our method outperforms strong baselines and achieves state-of-the-art performance on WikiTableQuestions and TabFact datasets.
Recent advances in motion planning for autonomous driving have led to models capable of generating high-quality trajectories. However, most existing planners tend to fix their policy after supervised training, leading to consistent but rigid driving behaviors. This limits their ability to reflect human preferences or adapt to dynamic, instruction-driven demands. In this work, we propose a diffusion-based multi-head trajectory planner(M-diffusion planner). During the early training stage, all output heads share weights to learn to generate high-quality trajectories. Leveraging the probabilistic nature of diffusion models, we then apply Group Relative Policy Optimization (GRPO) to fine-tune the pre-trained model for diverse policy-specific behaviors. At inference time, we incorporate a large language model (LLM) to guide strategy selection, enabling dynamic, instruction-aware planning without switching models. Closed-loop simulation demonstrates that our post-trained planner retains strong planning capability while achieving state-of-the-art (SOTA) performance on the nuPlan val14 benchmark. Open-loop results further show that the generated trajectories exhibit clear diversity, effectively satisfying multi-modal driving behavior requirements. The code and related experiments will be released upon acceptance of the paper.
We introduce PuzzleJAX, a GPU-accelerated puzzle game engine and description language designed to support rapid benchmarking of tree search, reinforcement learning, and LLM reasoning abilities. Unlike existing GPU-accelerated learning environments that provide hard-coded implementations of fixed sets of games, PuzzleJAX allows dynamic compilation of any game expressible in its domain-specific language (DSL). This DSL follows PuzzleScript, which is a popular and accessible online game engine for designing puzzle games. In this paper, we validate in PuzzleJAX several hundred of the thousands of games designed in PuzzleScript by both professional designers and casual creators since its release in 2013, thereby demonstrating PuzzleJAX's coverage of an expansive, expressive, and human-relevant space of tasks. By analyzing the performance of search, learning, and language models on these games, we show that PuzzleJAX can naturally express tasks that are both simple and intuitive to understand, yet often deeply challenging to master, requiring a combination of control, planning, and high-level insight.
Large language models (LLMs) have demonstrated remarkable capabilities across diverse domains, but their heavy resource demands make quantization-reducing precision to lower-bit formats-critical for efficient serving. While many quantization methods exist, a systematic understanding of their performance, energy, and quality tradeoffs in realistic serving conditions remains a gap. In this work, we first develop a fully automated online characterization framework qMeter, and then conduct an in-depth characterization of 11 post-training LLM quantization methods across 4 model sizes (7B-70B) and two GPU architectures (A100, H100). We evaluate quantization at the application, workload, parallelism, and hardware levels under online serving conditions. Our study reveals highly task- and method-dependent tradeoffs, strong sensitivity to workload characteristics, and complex interactions with parallelism and GPU architecture. We further present three optimization case studies illustrating deployment challenges in capacity planning, energy-efficient scheduling, and multi-objective tuning. To the best of our knowledge, this is one of the first comprehensive application-, system-, and hardware-level characterization of LLM quantization from a joint performance, energy, and quality perspective.
Recent advances in large language models (LLMs) and dense retrievers have driven significant progress in retrieval-augmented generation (RAG). However, existing approaches face significant challenges in complex reasoning-oriented multi-hop retrieval tasks: 1) Ineffective reasoning-oriented planning: Prior methods struggle to generate robust multi-step plans for complex queries, as rule-based decomposers perform poorly on out-of-template questions. 2) Suboptimal reasoning-driven retrieval: Related methods employ limited query reformulation, leading to iterative retrieval loops that often fail to locate golden documents. 3) Insufficient reasoning-guided filtering: Prevailing methods lack the fine-grained reasoning to effectively filter salient information from noisy results, hindering utilization of retrieved knowledge. Fundamentally, these limitations all stem from the weak coupling between retrieval and reasoning in current RAG architectures. We introduce the Orchestrated Planner-Executor Reasoning Architecture (OPERA), a novel reasoning-driven retrieval framework. OPERA's Goal Planning Module (GPM) decomposes questions into sub-goals, which are executed by a Reason-Execute Module (REM) with specialized components for precise reasoning and effective retrieval. To train OPERA, we propose Multi-Agents Progressive Group Relative Policy Optimization (MAPGRPO), a novel variant of GRPO. Experiments on complex multi-hop benchmarks show OPERA's superior performance, validating both the MAPGRPO method and OPERA's design. Code is available at https://github.com/Ameame1/OPERA.
Creating and improvising scenarios for content approaching is an enriching technique in education. However, it comes with a significant increase in the time spent on its planning, which intensifies when using complex technologies, such as social robots. Furthermore, addressing multicultural integration is commonly embedded in regular activities due to the already tight curriculum. Addressing these issues with a single solution, we implemented an intuitive interface that allows teachers to create scenario-based activities from their regular curriculum using LLMs and social robots. We co-designed different frameworks of activities with 4 teachers and deployed it in a study with 27 students for 1 week. Beyond validating the system's efficacy, our findings highlight the positive impact of integration policies perceived by the children and demonstrate the importance of scenario-based activities in students' enjoyment, observed to be significantly higher when applying storytelling. Additionally, several implications of using LLMs and social robots in long-term classroom activities are discussed.
With the development of Large Language Models (LLMs), numerous efforts have revealed their vulnerabilities to jailbreak attacks. Although these studies have driven the progress in LLMs' safety alignment, it remains unclear whether LLMs have internalized authentic knowledge to deal with real-world crimes, or are merely forced to simulate toxic language patterns. This ambiguity raises concerns that jailbreak success is often attributable to a hallucination loop between jailbroken LLM and judger LLM. By decoupling the use of jailbreak techniques, we construct knowledge-intensive Q\&A to investigate the misuse threats of LLMs in terms of dangerous knowledge possession, harmful task planning utility, and harmfulness judgment robustness. Experiments reveal a mismatch between jailbreak success rates and harmful knowledge possession in LLMs, and existing LLM-as-a-judge frameworks tend to anchor harmfulness judgments on toxic language patterns. Our study reveals a gap between existing LLM safety assessments and real-world threat potential.
Index recommendation is one of the most important problems in database management system (DBMS) optimization. Given queries and certain index-related constraints, traditional methods rely on heuristic optimization or learning-based models to select effective indexes and improve query performance. However, heuristic optimization suffers from high computation time, and learning-based models lose generalisability due to training for different workloads and database schemas. With the recent rapid development of large language models (LLMs), methods using prompt tuning have been proposed to enhance the efficiency of index selection. However, such methods still can not achieve the state-of-the-art (SOTA) results, and preparing the index selection demonstrations is also resource-intensive. To address these issues, we propose MAAdvisor, a zero-shot LLM-based index advisor with a multi-agent framework. We decompose the index recommendation problem into sub-steps, including planning, selection, combination, revision, and reflection. A set of LLM-embedded agents is designed to handle each one of the different sub-steps. Our method utilizes global agents to control the index selection process and local agents to select and revise indexes. Through extensive experiments, we show that our proposed MAAdvisor not only achieves the SOTA performance compared to the heuristic methods, but also outperforms learning-based and prompt-based methods with higher efficiency and better zero-shot inference ability.
Tool calling has emerged as a critical capability for AI agents to interact with the real world and solve complex tasks. While the Model Context Protocol (MCP) provides a powerful standardized framework for tool integration, there is a significant gap in benchmarking how well AI agents can effectively solve multi-step tasks using diverse MCP tools in realistic, dynamic scenarios. In this work, we present LiveMCP-101, a benchmark of 101 carefully curated real-world queries, refined through iterative LLM rewriting and manual review, that require coordinated use of multiple MCP tools including web search, file operations, mathematical reasoning, and data analysis. Moreover, we introduce a novel evaluation approach that leverages ground-truth execution plans rather than raw API outputs, better reflecting the evolving nature of real-world environments. Experiments show that even frontier LLMs achieve a success rate below 60\%, highlighting major challenges in tool orchestration. Detailed ablations and error analysis further reveal distinct failure modes and inefficiencies in token usage, pointing to concrete directions for advancing current models. LiveMCP-101 sets a rigorous standard for evaluating real-world agent capabilities, advancing toward autonomous AI systems that reliably execute complex tasks through tool use.
We introduce SRDrone, a novel system designed for self-refinement task planning in industrial-grade embodied drones. SRDrone incorporates two key technical contributions: First, it employs a continuous state evaluation methodology to robustly and accurately determine task outcomes and provide explanatory feedback. This approach supersedes conventional reliance on single-frame final-state assessment for continuous, dynamic drone operations. Second, SRDrone implements a hierarchical Behavior Tree (BT) modification model. This model integrates multi-level BT plan analysis with a constrained strategy space to enable structured reflective learning from experience. Experimental results demonstrate that SRDrone achieves a 44.87% improvement in Success Rate (SR) over baseline methods. Furthermore, real-world deployment utilizing an experience base optimized through iterative self-refinement attains a 96.25% SR. By embedding adaptive task refinement capabilities within an industrial-grade BT planning framework, SRDrone effectively integrates the general reasoning intelligence of Large Language Models (LLMs) with the stringent physical execution constraints inherent to embodied drones. Code is available at https://github.com/ZXiiiC/SRDrone.
We introduce Dream 7B, the most powerful open diffusion large language model to date. Unlike autoregressive (AR) models that generate tokens sequentially, Dream 7B employs discrete diffusion modeling to refine sequences in parallel through iterative denoising. Our model consistently outperforms existing diffusion language models on general, mathematical, and coding tasks. Dream 7B demonstrates superior planning abilities and inference flexibility, including arbitrary-order generation, infilling capabilities, and tunable quality-speed trade-offs. These results are achieved through simple yet effective training techniques, including AR-based LLM initialization and context-adaptive token-level noise rescheduling. We release both Dream-Base and Dream-Instruct to facilitate further research in diffusion-based language modeling.