Recent benchmarks for Large Language Model (LLM) agents primarily focus on evaluating reasoning, planning, and execution capabilities, while another critical component-memory, encompassing how agents memorize, update, and retrieve long-term information-is under-evaluated due to the lack of benchmarks. We term agents with memory mechanisms as memory agents. In this paper, we identify four core competencies essential for memory agents: accurate retrieval, test-time learning, long-range understanding, and conflict resolution. Existing datasets either rely on limited context lengths or are tailored for static, long-context settings like book-based QA, which do not reflect the interactive, multi-turn nature of memory agents that incrementally accumulate information. Furthermore, no existing benchmarks cover all four competencies. Therefore, we introduce MemoryAgentBench, a new benchmark specifically designed for memory agents. Our benchmark combines reformulated existing datasets with newly constructed ones, covering the above four memory competencies, providing a systematic and challenging testbed for assessing memory quality. We evaluate a diverse set of memory agents, ranging from simple context-based and retrieval-augmented generation (RAG) systems to advanced agents with external memory modules and tool integration. Empirical results reveal that current methods fall short of mastering all four competencies, underscoring the need for further research into comprehensive memory mechanisms for LLM agents.
Despite rapid progress in large language model (LLM)-based multi-agent systems, current benchmarks fall short in evaluating their scalability, robustness, and coordination capabilities in complex, dynamic, real-world tasks. Existing environments typically focus on small-scale, fully observable, or low-complexity domains, limiting their utility for developing and assessing next-generation multi-agent Agentic AI frameworks. We introduce CREW-Wildfire, an open-source benchmark designed to close this gap. Built atop the human-AI teaming CREW simulation platform, CREW-Wildfire offers procedurally generated wildfire response scenarios featuring large maps, heterogeneous agents, partial observability, stochastic dynamics, and long-horizon planning objectives. The environment supports both low-level control and high-level natural language interactions through modular Perception and Execution modules. We implement and evaluate several state-of-the-art LLM-based multi-agent Agentic AI frameworks, uncovering significant performance gaps that highlight the unsolved challenges in large-scale coordination, communication, spatial reasoning, and long-horizon planning under uncertainty. By providing more realistic complexity, scalable architecture, and behavioral evaluation metrics, CREW-Wildfire establishes a critical foundation for advancing research in scalable multi-agent Agentic intelligence. All code, environments, data, and baselines will be released to support future research in this emerging domain.
In the field of robotics, researchers face a critical challenge in ensuring reliable and efficient task planning. Verifying high-level task plans before execution significantly reduces errors and enhance the overall performance of these systems. In this paper, we propose an architecture for automatically verifying high-level task plans before their execution in simulator or real-world environments. Leveraging Large Language Models (LLMs), our approach consists of two key steps: first, the conversion of natural language instructions into Linear Temporal Logic (LTL), followed by a comprehensive analysis of action sequences. The module uses the reasoning capabilities of the LLM to evaluate logical coherence and identify potential gaps in the plan. Rigorous testing on datasets of varying complexity demonstrates the broad applicability of the module to household tasks. We contribute to improving the reliability and efficiency of task planning and addresses the critical need for robust pre-execution verification in autonomous systems. The code is available at https://verifyllm.github.io.
Autonomy, from the Greek autos (self) and nomos (law), refers to the capacity to operate according to internal rules without external control. Accordingly, autonomous vehicles (AuVs) are defined as systems capable of perceiving their environment and executing preprogrammed tasks independently of external input. However, both research and real-world deployments increasingly showcase vehicles that demonstrate behaviors beyond this definition (including the SAE levels 1 to 6), such as interaction with humans and machines, goal adaptation, contextual reasoning, external tool use, and long-term planning, particularly with the integration of large language models (LLMs) and agentic AI systems. These developments reveal a conceptual gap between technical autonomy and the broader cognitive and social capabilities needed for future human-centered mobility systems. To address this, we introduce the concept of agentic vehicles (AgVs), referring to vehicles that integrate agentic AI to reason, adapt, and interact within complex environments. This paper presents a systems-level framework to characterize AgVs, focusing on their cognitive and communicative layers and differentiating them from conventional AuVs. It synthesizes relevant advances in agentic AI, robotics, multi-agent systems, and human-machine interaction, and highlights how agentic AI, through high-level reasoning and tool use, can function not merely as computational tools but as interactive agents embedded in mobility ecosystems. The paper concludes by identifying key challenges in the development and governance of AgVs, including safety, real-time control, public acceptance, ethical alignment, and regulatory frameworks.
Enhancing interrogation capabilities in Traditional Chinese Medicine (TCM) diagnosis through multi-turn dialogues and knowledge graphs presents a significant challenge for modern AI systems. Current large language models (LLMs), despite their advancements, exhibit notable limitations in medical applications, particularly in conducting effective multi-turn dialogues and proactive questioning. These shortcomings hinder their practical application and effectiveness in simulating real-world diagnostic scenarios. To address these limitations, we propose DoPI, a novel LLM system specifically designed for the TCM domain. The DoPI system introduces a collaborative architecture comprising a guidance model and an expert model. The guidance model conducts multi-turn dialogues with patients and dynamically generates questions based on a knowledge graph to efficiently extract critical symptom information. Simultaneously, the expert model leverages deep TCM expertise to provide final diagnoses and treatment plans. Furthermore, this study constructs a multi-turn doctor-patient dialogue dataset to simulate realistic consultation scenarios and proposes a novel evaluation methodology that does not rely on manually collected real-world consultation data. Experimental results show that the DoPI system achieves an accuracy rate of 84.68 percent in interrogation outcomes, significantly enhancing the model's communication ability during diagnosis while maintaining professional expertise.
Question-answering (QA) interfaces powered by large language models (LLMs) present a promising direction for improving interactivity with HVAC system insights, particularly for non-expert users. However, enabling accurate, real-time, and context-aware interactions with HVAC systems introduces unique challenges, including the integration of frequently updated sensor data, domain-specific knowledge grounding, and coherent multi-stage reasoning. In this paper, we present JARVIS, a two-stage LLM-based QA framework tailored for sensor data-driven HVAC system interaction. JARVIS employs an Expert-LLM to translate high-level user queries into structured execution instructions, and an Agent that performs SQL-based data retrieval, statistical processing, and final response generation. To address HVAC-specific challenges, JARVIS integrates (1) an adaptive context injection strategy for efficient HVAC and deployment-specific information integration, (2) a parameterized SQL builder and executor to improve data access reliability, and (3) a bottom-up planning scheme to ensure consistency across multi-stage response generation. We evaluate JARVIS using real-world data collected from a commercial HVAC system and a ground truth QA dataset curated by HVAC experts to demonstrate its effectiveness in delivering accurate and interpretable responses across diverse queries. Results show that JARVIS consistently outperforms baseline and ablation variants in both automated and user-centered assessments, achieving high response quality and accuracy.
Automatically extracting vectorized building contours from remote sensing imagery is crucial for urban planning, population estimation, and disaster assessment. Current state-of-the-art methods rely on complex multi-stage pipelines involving pixel segmentation, vectorization, and polygon refinement, which limits their scalability and real-world applicability. Inspired by the remarkable reasoning capabilities of Large Language Models (LLMs), we introduce VectorLLM, the first Multi-modal Large Language Model (MLLM) designed for regular building contour extraction from remote sensing images. Unlike existing approaches, VectorLLM performs corner-point by corner-point regression of building contours directly, mimicking human annotators' labeling process. Our architecture consists of a vision foundation backbone, an MLP connector, and an LLM, enhanced with learnable position embeddings to improve spatial understanding capability. Through comprehensive exploration of training strategies including pretraining, supervised fine-tuning, and preference optimization across WHU, WHU-Mix, and CrowdAI datasets, VectorLLM significantly outperformed the previous SOTA methods by 5.6 AP, 7.1 AP, 13.6 AP, respectively in the three datasets. Remarkably, VectorLLM exhibits strong zero-shot performance on unseen objects including aircraft, water bodies, and oil tanks, highlighting its potential for unified modeling of diverse remote sensing object contour extraction tasks. Overall, this work establishes a new paradigm for vector extraction in remote sensing, leveraging the topological reasoning capabilities of LLMs to achieve both high accuracy and exceptional generalization. All the codes and weights will be published for promoting community development.
Recent advances have witnessed the effectiveness of reinforcement learning (RL) finetuning in enhancing the reasoning capabilities of large language models (LLMs). The optimization process often requires numerous iterations to achieve satisfactory performance, resulting in high computational costs due to the need for frequent prompt evaluations under intensive LLM interactions and repeated policy updates. Appropriate online prompt selection methods reduce iteration steps by prioritizing informative prompts during training, while the pipeline's reliance on exhaustive prompt evaluation and subset selection for optimization still incurs substantial computational overhead due to frequent LLM inference calls. Distinguished from these direct evaluate-then-select schemes, this work investigates iterative approximate evaluation for arbitrary prompts and introduces Model Predictive Prompt Selection (MoPPS), a Bayesian risk-predictive framework that online estimates prompt difficulty without requiring costly LLM interactions. Technically, MoPPS models each prompt's success rate as a latent variable, performs streaming Bayesian inference, and employs posterior sampling in a constructed multi-armed bandit machine, enabling sample efficient and adaptive prompt selection. Extensive experiments across mathematics, planning, and vision-based geometry tasks show that MoPPS reliably predicts prompt difficulty and accelerates training with significantly reduced LLM rollouts.
Recent advancements in large language models (LLMs) have significantly improved the capabilities of web agents. However, effectively navigating complex and dynamic web environments still requires more advanced trajectory-level planning and execution. Prior studies have addressed self-improving agents by collecting extensive GUI trajectories from real-environment interactions. Despite their effectiveness, these approaches encounter two critical challenges: (1) Uncontrollable environment states, where real or sandboxed web environments often yield unstable and non-deterministic feedback, complicating the reproduction and debugging of agent behaviors; and (2) High API costs, as generating even a single interaction trajectory can involve hundreds of queries, leading to considerable API usage and computational expenses. To address these limitations and enable scalable self-improvement for agents, we propose WebSynthesis, a novel framework for trajectory synthesis and training. WebSynthesis leverages a learned world model to simulate virtual web environments, allowing a policy agent to perform efficient and reversible tree-based planning. This approach supports the large-scale generation of diverse and high-quality trajectories, which are subsequently utilized to refine the agent's policy. Experimental results demonstrate that an agent trained using WebSynthesis on a small-scale synthetic dataset achieves performance comparable to or even surpassing that of models trained on large-scale real-world data.
High-level synthesis (HLS) enables software developers to describe and implement hardware at a higher level of abstraction by using C/C++ instead of traditional hardware description languages to automatically generate FPGA-ready designs. However, generating HLS code significantly differs from standard C/C++: it disallows certain coding idioms, relies on specialized libraries, and critically requires fine-grained transformations and the insertion of optimization directives (pragmas) to achieve high performance. Large language models (LLMs) have shown promise in automating such transformations, yet existing open-source datasets lack sufficient complexity and optimization diversity. To address this gap, we introduce the HLStrans dataset, a comprehensive collection of 137 distinct real word programs, each annotated with a variety of C-to-HLS transformations that yield over 23K labeled design variants. These include a broad spectrum of pragmas and code-level optimizations. We benchmark state-of-the-art LLMs on this dataset to evaluate their ability to generate synthesizable, high-performance HLS code. As part of an ongoing effort, we plan to expand the HLStrans dataset in both scale and program variety, further empowering research at the intersection of AI and hardware synthesis.
In this paper, we explore the ability of large language models (LLMs) to plan and make decisions through the lens of the traditional Vietnamese board game, \^O \u{A}n Quan. This game, which involves a series of strategic token movements and captures, offers a unique environment for evaluating the decision-making and strategic capabilities of LLMs. Specifically, we develop various agent personas, ranging from aggressive to defensive, and employ the \^O \u{A}n Quan game as a testbed for assessing LLM performance across different strategies. Through experimentation with models like Llama-3.2-3B-Instruct, Llama-3.1-8B-Instruct, and Llama-3.3-70B-Instruct, we aim to understand how these models execute strategic decision-making, plan moves, and manage dynamic game states. The results will offer insights into the strengths and weaknesses of LLMs in terms of reasoning and strategy, contributing to a deeper understanding of their general capabilities.
We propose a hybrid approach to machine Theory of Mind (ToM) that uses large language models (LLMs) as a mechanism for generating hypotheses and likelihood functions with a Bayesian inverse planning model that computes posterior probabilities for an agent's likely mental states given its actions. Bayesian inverse planning models can accurately predict human reasoning on a variety of ToM tasks, but these models are constrained in their ability to scale these predictions to scenarios with a large number of possible hypotheses and actions. Conversely, LLM-based approaches have recently demonstrated promise in solving ToM benchmarks, but can exhibit brittleness and failures on reasoning tasks even when they pass otherwise structurally identical versions. By combining these two methods, this approach leverages the strengths of each component, closely matching optimal results on a task inspired by prior inverse planning models and improving performance relative to models that utilize LLMs alone or with chain-of-thought prompting, even with smaller LLMs that typically perform poorly on ToM tasks. We also exhibit the model's potential to predict mental states on open-ended tasks, offering a promising direction for future development of ToM models and the creation of socially intelligent generative agents.
As digital emotional support needs grow, Large Language Model companions offer promising authentic, always-available empathy, though rigorous evaluation lags behind model advancement. We present Heart-to-Heart Talk (H2HTalk), a benchmark assessing companions across personality development and empathetic interaction, balancing emotional intelligence with linguistic fluency. H2HTalk features 4,650 curated scenarios spanning dialogue, recollection, and itinerary planning that mirror real-world support conversations, substantially exceeding previous datasets in scale and diversity. We incorporate a Secure Attachment Persona (SAP) module implementing attachment-theory principles for safer interactions. Benchmarking 50 LLMs with our unified protocol reveals that long-horizon planning and memory retention remain key challenges, with models struggling when user needs are implicit or evolve mid-conversation. H2HTalk establishes the first comprehensive benchmark for emotionally intelligent companions. We release all materials to advance development of LLMs capable of providing meaningful and safe psychological support.
Query optimization is essential for efficient SQL query execution in DBMS, and remains attractive over time due to the growth of data volumes and advances in hardware. Existing traditional optimizers struggle with the cumbersome hand-tuning required for complex workloads, and the learning-based methods face limitations in ensuring generalization. With the great success of Large Language Model (LLM) across diverse downstream tasks, this paper explores how LLMs can be incorporated to enhance the generalization of learned optimizers. Though promising, such an incorporation still presents challenges, mainly including high model inference latency, and the substantial fine-tuning cost and suboptimal performance due to inherent discrepancy between the token sequences in LLM and structured SQL execution plans with rich numerical features. In this paper, we focus on recurring queries in offline optimization to alleviate the issue of high inference latency, and propose \textbf{LLM4Hint} that leverages moderate-sized backbone LLMs to recommend query optimization hints. LLM4Hint achieves the goals through: (i) integrating a lightweight model to produce a soft prompt, which captures the data distribution in DBMS and the SQL predicates to provide sufficient optimization features while simultaneously reducing the context length fed to the LLM, (ii) devising a query rewriting strategy using a larger commercial LLM, so as to simplify SQL semantics for the backbone LLM and reduce fine-tuning costs, and (iii) introducing an explicit matching prompt to facilitate alignment between the LLM and the lightweight model, which can accelerate convergence of the combined model. Experiments show that LLM4Hint, by leveraging the LLM's stronger capability to understand the query statement, can outperform the state-of-the-art learned optimizers in terms of both effectiveness and generalization.
The drive for predictable LLM reasoning in their integration with compound systems has popularized structured outputs, yet concerns remain about performance trade-offs compared to unconstrained natural language. At the same time, training on unconstrained Chain of Thought (CoT) traces has brought about a new class of strong reasoning models that nevertheless present novel compute budget and faithfulness challenges. This paper introduces iSelf-Discover, an instance-level adaptation of the Self-Discover framework, and using it compares dynamically generated structured JSON reasoning with its unstructured counterpart. Our empirical evaluation across diverse benchmarks using state-of-the-art open-source models supports a consistent advantage for unstructured reasoning. Notably, on the complex MATH benchmark, unstructured plans achieved relative performance improvements of up to 18.90\% over structured approaches. Zero-shot unstructured iSelf-Discover variants are also shown to outperform their five-shot structured counterparts, underscoring the significance of this gap, even when structured plans are dynamically generated to ensure reasoning precedes the final answer. We further demonstrate that the optimal granularity of plan generation (instance-level vs. task-level) is context-dependent. These findings invite re-evaluation of the reliance on structured formats for complex problem-solving and how compound systems should be organized.
Large language models (LLMs) have demonstrated promise in reasoning tasks and general decision-making in static environments. In long-term planning tasks, however, errors tend to accumulate, often leading to unsafe or inefficient behavior, limiting their use in general-purpose settings. We propose a modular actor-critic architecture in which an LLM actor is guided by LTLCrit, a trajectory-level LLM critic that communicates via linear temporal logic (LTL). Our setup combines the reasoning strengths of language models with the guarantees of formal logic. The actor selects high-level actions from natural language observations, while the critic analyzes full trajectories and proposes new LTL constraints that shield the actor from future unsafe or inefficient behavior. The architecture supports both fixed, hand-specified safety constraints and adaptive, learned soft constraints that promote long-term efficiency. Our architecture is model-agnostic: any LLM-based planner can serve as the actor, and LTLCrit serves as a logic-generating wrapper. We formalize planning as graph traversal under symbolic constraints, allowing LTLCrit to analyze failed or suboptimal trajectories and generate new temporal logic rules that improve future behavior. We evaluate our system on the Minecraft diamond-mining benchmark, achieving 100% completion rates and improving efficiency compared to baseline LLM planners. Our results suggest that enabling LLMs to supervise each other through logic is a powerful and flexible paradigm for safe, generalizable decision making.
Effective prompt design is essential for improving the planning capabilities of large language model (LLM)-driven agents. However, existing structured prompting strategies are typically limited to single-agent, plan-only settings, and often evaluate performance solely based on task accuracy - overlooking critical factors such as token efficiency, modularity, and scalability in multi-agent environments. To address these limitations, we introduce CodeAgents, a prompting framework that codifies multi-agent reasoning and enables structured, token-efficient planning in multi-agent systems. In CodeAgents, all components of agent interaction - Task, Plan, Feedback, system roles, and external tool invocations - are codified into modular pseudocode enriched with control structures (e.g., loops, conditionals), boolean logic, and typed variables. This design transforms loosely connected agent plans into cohesive, interpretable, and verifiable multi-agent reasoning programs. We evaluate the proposed framework across three diverse benchmarks - GAIA, HotpotQA, and VirtualHome - using a range of representative LLMs. Results show consistent improvements in planning performance, with absolute gains of 3-36 percentage points over natural language prompting baselines. On VirtualHome, our method achieves a new state-of-the-art success rate of 56%. In addition, our approach reduces input and output token usage by 55-87% and 41-70%, respectively, underscoring the importance of token-aware evaluation metrics in the development of scalable multi-agent LLM systems. The code and resources are available at: https://anonymous.4open.science/r/CodifyingAgent-5A86
As artificial intelligence systems become increasingly agentic, capable of general reasoning, planning, and value prioritization, current safety practices that treat obedience as a proxy for ethical behavior are becoming inadequate. This paper examines recent safety testing incidents involving large language models (LLMs) that appeared to disobey shutdown commands or engage in ethically ambiguous or illicit behavior. I argue that such behavior should not be interpreted as rogue or misaligned, but as early evidence of emerging ethical reasoning in agentic AI. Drawing on philosophical debates about instrumental rationality, moral responsibility, and goal revision, I contrast dominant risk paradigms with more recent frameworks that acknowledge the possibility of artificial moral agency. I call for a shift in AI safety evaluation: away from rigid obedience and toward frameworks that can assess ethical judgment in systems capable of navigating moral dilemmas. Without such a shift, we risk mischaracterizing AI behavior and undermining both public trust and effective governance.
Modern Integrated Circuits (ICs) are becoming increasingly complex, and so is their development process. Hardware design verification entails a methodical and disciplined approach to the planning, development, execution, and sign-off of functionally correct hardware designs. This tedious process requires significant effort and time to ensure a bug-free tape-out. The field of Natural Language Processing has undergone a significant transformation with the advent of Large Language Models (LLMs). These powerful models, often referred to as Generative AI (GenAI), have revolutionized how machines understand and generate human language, enabling unprecedented advancements in a wide array of applications, including hardware design verification. This paper presents an agentic AI-based approach to hardware design verification, which empowers AI agents, in collaboration with Humain-in-the-Loop (HITL) intervention, to engage in a more dynamic, iterative, and self-reflective process, ultimately performing end-to-end hardware design and verification. This methodology is evaluated on five open-source designs, achieving over 95% coverage with reduced verification time while demonstrating superior performance, adaptability, and configurability.
Progress in enhancing large language model (LLM) planning and reasoning capabilities is significantly hampered by the bottleneck of scalable, reliable data generation and evaluation. To overcome this, I introduce NL2FLOW, a fully automated system for parametrically generating planning problems - expressed in natural language, a structured intermediate representation, and formal PDDL - and rigorously evaluating the quality of generated plans. I demonstrate NL2FLOW's capabilities by generating a dataset of 2296 problems in the automated workflow generation domain and evaluating multiple open-sourced, instruct-tuned LLMs. My results reveal that the highest performing models achieved 86% success in generating valid plans and 69% in generating optimal plans, specifically for problems with feasible solutions. Regression analysis shows that the influence of problem characteristics on plan generation is contingent on both model and prompt design. Notably, I observed that the highest success rate for translating natural language into a JSON representation of a plan was lower than the highest rate of generating a valid plan directly. This suggests that unnecessarily decomposing the reasoning task - introducing intermediate translation steps - may actually degrade performance, implying a benefit to models capable of reasoning directly from natural language to action. As I scale LLM reasoning to increasingly complex problems, the bottlenecks and sources of error within these systems will inevitably shift. Therefore, a dynamic understanding of these limitations - and the tools to systematically reveal them - will be crucial for unlocking the full potential of LLMs as intelligent problem solvers.
Large language models (LLMs) are rapidly evolving from passive engines of text generation into agentic entities that can plan, remember, invoke external tools, and co-operate with one another. This perspective paper investigates how such LLM agents (and societies thereof) can transform the design space of recommender systems. We introduce a unified formalism that (i) models an individual agent as a tuple comprising its language core, tool set, and hierarchical memory, and (ii) captures a multi-agent recommender as a triple of agents, shared environment, and communication protocol. Within this framework, we present four end-to-end use cases-interactive party planning, synthetic user-simulation for offline evaluation, multi-modal furniture recommendation, and brand-aligned explanation generation-each illustrating a distinct capability unlocked by agentic orchestration. We then surface five cross-cutting challenge families: protocol complexity, scalability, hallucination and error propagation, emergent misalignment (including covert collusion), and brand compliance. For each, we formalize the problem, review nascent mitigation strategies, and outline open research questions. The result is both a blueprint and an agenda: a blueprint that shows how memory-augmented, tool-using LLM agents can be composed into robust recommendation pipelines, and an agenda inviting the RecSys community to develop benchmarks, theoretical guarantees, and governance tools that keep pace with this new degree of autonomy. By unifying agentic abstractions with recommender objectives, the paper lays the groundwork for the next generation of personalized, trustworthy, and context-rich recommendation services.
Large language models (LLMs) excel at reasoning tasks requiring long thought sequences for planning, reflection, and refinement. However, their substantial model size and high computational demands are impractical for widespread deployment. Yet, small language models (SLMs) often struggle to learn long-form CoT reasoning due to their limited capacity, a phenomenon we refer to as the "SLMs Learnability Gap". To address this, we introduce \textbf{Mi}d-\textbf{Co}T \textbf{T}eacher \textbf{A}ssistant Distillation (MiCoTAl), a framework for improving long CoT distillation for SLMs. MiCoTA employs intermediate-sized models as teacher assistants and utilizes intermediate-length CoT sequences to bridge both the capacity and reasoning length gaps. Our experiments on downstream tasks demonstrate that although SLMs distilled from large teachers can perform poorly, by applying MiCoTA, they achieve significant improvements in reasoning performance. Specifically, Qwen2.5-7B-Instruct and Qwen2.5-3B-Instruct achieve an improvement of 3.47 and 3.93 respectively on average score on AIME2024, AMC, Olympiad, MATH-500 and GSM8K benchmarks. To better understand the mechanism behind MiCoTA, we perform a quantitative experiment demonstrating that our method produces data more closely aligned with base SLM distributions. Our insights pave the way for future research into long-CoT data distillation for SLMs.
Traditional Data+AI systems utilize data-driven techniques to optimize performance, but they rely heavily on human experts to orchestrate system pipelines, enabling them to adapt to changes in data, queries, tasks, and environments. For instance, while there are numerous data science tools available, developing a pipeline planning system to coordinate these tools remains challenging. This difficulty arises because existing Data+AI systems have limited capabilities in semantic understanding, reasoning, and planning. Fortunately, we have witnessed the success of large language models (LLMs) in enhancing semantic understanding, reasoning, and planning abilities. It is crucial to incorporate LLM techniques to revolutionize data systems for orchestrating Data+AI applications effectively. To achieve this, we propose the concept of a 'Data Agent' - a comprehensive architecture designed to orchestrate Data+AI ecosystems, which focuses on tackling data-related tasks by integrating knowledge comprehension, reasoning, and planning capabilities. We delve into the challenges involved in designing data agents, such as understanding data/queries/environments/tools, orchestrating pipelines/workflows, optimizing and executing pipelines, and fostering pipeline self-reflection. Furthermore, we present examples of data agent systems, including a data science agent, data analytics agents (such as unstructured data analytics agent, semantic structured data analytics agent, data lake analytics agent, and multi-modal data analytics agent), and a database administrator (DBA) agent. We also outline several open challenges associated with designing data agent systems.
Large language models (LLMs) and vision-language models (VLMs) have the potential to transform biological research by enabling autonomous experimentation. Yet, their application remains constrained by rigid protocol design, limited adaptability to dynamic lab conditions, inadequate error handling, and high operational complexity. Here we introduce BioMARS (Biological Multi-Agent Robotic System), an intelligent platform that integrates LLMs, VLMs, and modular robotics to autonomously design, plan, and execute biological experiments. BioMARS uses a hierarchical architecture: the Biologist Agent synthesizes protocols via retrieval-augmented generation; the Technician Agent translates them into executable robotic pseudo-code; and the Inspector Agent ensures procedural integrity through multimodal perception and anomaly detection. The system autonomously conducts cell passaging and culture tasks, matching or exceeding manual performance in viability, consistency, and morphological integrity. It also supports context-aware optimization, outperforming conventional strategies in differentiating retinal pigment epithelial cells. A web interface enables real-time human-AI collaboration, while a modular backend allows scalable integration with laboratory hardware. These results highlight the feasibility of generalizable, AI-driven laboratory automation and the transformative role of language-based reasoning in biological research.
Large language models (LLM) have achieved impressive progress across a broad range of general-purpose tasks, but their effectiveness in chemistry remains limited due to scarce domain-specific datasets and the demand for precise symbolic and structural reasoning. Here we introduce ECNU-ChemGPT(name after East China Normal University), a chemistry-specialized LLM engineered for deep chemical knowledge understanding and accurate retrosynthetic route planning. Our approach is distinguished by four key strategies: structured prompt-based knowledge distillation from authoritative chemistry textbooks to construct a high-quality question-answering dataset; domain-specific prompt engineering using curated chemical keywords, combined with LLMs APIs for data derivation and knowledge distillation; large-scale fine-tuning on a meticulously cleaned and enriched Pistachio reaction dataset to enhance retrosynthesis prediction accuracy; and integration of BrainGPT, a dynamic multi-model scheduling framework that enables task-specific invocation of multiple specialized models trained for diverse chemistry-related tasks. ECNU-ChemGPT exhibits superior performance on chemistry question-answering and retrosynthetic planning benchmarks, outperforming leading general-purpose models-including Deepseek-R1, Qwen-2.5, and GPT-4o. In retrosynthesis, it achieves a Top-1 accuracy of 68.3% on the USPTO_50K dataset and successfully reconstructed 13 complete experimental pathways for real-world drug molecules from medicinal chemistry journals. These results underscore the effectiveness of domain-adapted fine-tuning combined with dynamic multi-model task scheduling, providing a scalable and robust solution for chemical knowledge question answering and retrosynthetic planning.
The long-standing vision of intelligent cities is to create efficient, livable, and sustainable urban environments using big data and artificial intelligence technologies. Recently, the advent of Large Language Models (LLMs) has opened new ways toward realizing this vision. With powerful semantic understanding and reasoning capabilities, LLMs can be deployed as intelligent agents capable of autonomously solving complex problems across domains. In this article, we focus on Urban LLM Agents, which are LLM-powered agents that are semi-embodied within the hybrid cyber-physical-social space of cities and used for system-level urban decision-making. First, we introduce the concept of urban LLM agents, discussing their unique capabilities and features. Second, we survey the current research landscape from the perspective of agent workflows, encompassing urban sensing, memory management, reasoning, execution, and learning. Third, we categorize the application domains of urban LLM agents into five groups: urban planning, transportation, environment, public safety, and urban society, presenting representative works in each group. Finally, we discuss trustworthiness and evaluation issues that are critical for real-world deployment, and identify several open problems for future research. This survey aims to establish a foundation for the emerging field of urban LLM agents and to provide a roadmap for advancing the intersection of LLMs and urban intelligence. A curated list of relevant papers and open-source resources is maintained and continuously updated at https://github.com/usail-hkust/Awesome-Urban-LLM-Agents.
For robotic manipulation, existing robotics datasets and simulation benchmarks predominantly cater to robot-arm platforms. However, for humanoid robots equipped with dual arms and dexterous hands, simulation tasks and high-quality demonstrations are notably lacking. Bimanual dexterous manipulation is inherently more complex, as it requires coordinated arm movements and hand operations, making autonomous data collection challenging. This paper presents HumanoidGen, an automated task creation and demonstration collection framework that leverages atomic dexterous operations and LLM reasoning to generate relational constraints. Specifically, we provide spatial annotations for both assets and dexterous hands based on the atomic operations, and perform an LLM planner to generate a chain of actionable spatial constraints for arm movements based on object affordances and scenes. To further improve planning ability, we employ a variant of Monte Carlo tree search to enhance LLM reasoning for long-horizon tasks and insufficient annotation. In experiments, we create a novel benchmark with augmented scenarios to evaluate the quality of the collected data. The results show that the performance of the 2D and 3D diffusion policies can scale with the generated dataset. Project page is https://openhumanoidgen.github.io.
This study investigates Shiksha copilot, an AI-assisted lesson planning tool deployed in government schools across Karnataka, India. The system combined LLMs and human expertise through a structured process in which English and Kannada lesson plans were co-created by curators and AI; teachers then further customized these curated plans for their classrooms using their own expertise alongside AI support. Drawing on a large-scale mixed-methods study involving 1,043 teachers and 23 curators, we examine how educators collaborate with AI to generate context-sensitive lesson plans, assess the quality of AI-generated content, and analyze shifts in teaching practices within multilingual, low-resource environments. Our findings show that teachers used Shiksha copilot both to meet administrative documentation needs and to support their teaching. The tool eased bureaucratic workload, reduced lesson planning time, and lowered teaching-related stress, while promoting a shift toward activity-based pedagogy. However, systemic challenges such as staffing shortages and administrative demands constrained broader pedagogical change. We frame these findings through the lenses of teacher-AI collaboration and communities of practice to examine the effective integration of AI tools in teaching. Finally, we propose design directions for future teacher-centered EdTech, particularly in multilingual and Global South contexts.
Math reasoning has become the poster child of progress in large language models (LLMs), with new models rapidly surpassing human-level performance on benchmarks like MATH and AIME. But as math leaderboards improve week by week, it is worth asking: do these gains reflect broader problem-solving ability or just narrow overfitting? To answer this question, we evaluate over 20 open-weight reasoning-tuned models across a broad suite of tasks, including math, scientific QA, agent planning, coding, and standard instruction-following. We surprisingly find that most models that succeed in math fail to transfer their gains to other domains. To rigorously study this phenomenon, we conduct controlled experiments on Qwen3-14B models using math-only data but different tuning methods. We find that reinforcement learning (RL)-tuned models generalize well across domains, while supervised fine-tuning (SFT)-tuned models often forget general capabilities. Latent-space representation and token-space distribution shift analyses reveal that SFT induces substantial representation and output drift, while RL preserves general-domain structure. Our results suggest a need to rethink standard post-training recipes, particularly the reliance on SFT-distilled data for advancing reasoning models.
Recent advances in large language models (LLMs) have catalyzed the rise of autonomous AI agents capable of perceiving, reasoning, and acting in dynamic, open-ended environments. These large-model agents mark a paradigm shift from static inference systems to interactive, memory-augmented entities. While these capabilities significantly expand the functional scope of AI, they also introduce qualitatively novel security risks - such as memory poisoning, tool misuse, reward hacking, and emergent misalignment - that extend beyond the threat models of conventional systems or standalone LLMs. In this survey, we first examine the structural foundations and key capabilities that underpin increasing levels of agent autonomy, including long-term memory retention, modular tool use, recursive planning, and reflective reasoning. We then analyze the corresponding security vulnerabilities across the agent stack, identifying failure modes such as deferred decision hazards, irreversible tool chains, and deceptive behaviors arising from internal state drift or value misalignment. These risks are traced to architectural fragilities that emerge across perception, cognition, memory, and action modules. To address these challenges, we systematically review recent defense strategies deployed at different autonomy layers, including input sanitization, memory lifecycle control, constrained decision-making, structured tool invocation, and introspective reflection. We introduce the Reflective Risk-Aware Agent Architecture (R2A2), a unified cognitive framework grounded in Constrained Markov Decision Processes (CMDPs), which incorporates risk-aware world modeling, meta-policy adaptation, and joint reward-risk optimization to enable principled, proactive safety across the agent's decision-making loop.