Large language models (LLMs) have shown strong reasoning and coding capabilities, yet they struggle to generalize to real-world software engineering (SWE) problems that are long-horizon and out of distribution. Existing systems often rely on a single agent to handle the entire workflow-interpreting issues, navigating large codebases, and implementing fixes-within one reasoning chain. Such monolithic designs force the model to retain irrelevant context, leading to spurious correlations and poor generalization. Motivated by how human engineers decompose complex problems, we propose structuring SWE agents as orchestrators coordinating specialized sub-agents for sub-tasks such as localization, editing, and validation. The challenge lies in discovering effective hierarchies automatically: as the number of sub-agents grows, the search space becomes combinatorial, and it is difficult to attribute credit to individual sub-agents within a team. We address these challenges by formulating hierarchy discovery as a multi-armed bandit (MAB) problem, where each arm represents a candidate sub-agent and the reward measures its helpfulness when collaborating with others. This framework, termed Bandit Optimization for Agent Design (BOAD), enables efficient exploration of sub-agent designs under limited evaluation budgets. On SWE-bench-Verified, BOAD outperforms single-agent and manually designed multi-agent systems. On SWE-bench-Live, featuring more recent and out-of-distribution issues, our 36B system ranks second on the leaderboard at the time of evaluation, surpassing larger models such as GPT-4 and Claude. These results demonstrate that automatically discovered hierarchical multi-agent systems significantly improve generalization on challenging long-horizon SWE tasks. Code is available at https://github.com/iamxjy/BOAD-SWE-Agent.
Enabling Large Language Models (LLMs) to reliably invoke external tools remains a critical bottleneck for autonomous agents. Existing approaches suffer from three fundamental challenges: expensive human annotation for high-quality trajectories, poor generalization to unseen tools, and quality ceilings inherent in single-model synthesis that perpetuate biases and coverage gaps. We introduce InfTool, a fully autonomous framework that breaks these barriers through self-evolving multi-agent synthesis. Given only raw API specifications, InfTool orchestrates three collaborative agents (User Simulator, Tool-Calling Assistant, and MCP Server) to generate diverse, verified trajectories spanning single-turn calls to complex multi-step workflows. The framework establishes a closed loop: synthesized data trains the model via Group Relative Policy Optimization (GRPO) with gated rewards, the improved model generates higher-quality data targeting capability gaps, and this cycle iterates without human intervention. Experiments on the Berkeley Function-Calling Leaderboard (BFCL) demonstrate that InfTool transforms a base 32B model from 19.8% to 70.9% accuracy (+258%), surpassing models 10x larger and rivaling Claude-Opus, and entirely from synthetic data without human annotation.
Powerful autonomous systems, which reason, plan, and converse using and between numerous tools and agents, are made possible by Large Language Models (LLMs), Vision-Language Models (VLMs), and new agentic AI systems, like LangChain and GraphChain. Nevertheless, this agentic environment increases the probability of the occurrence of multimodal prompt injection (PI) attacks, in which concealed or malicious instructions carried in text, pictures, metadata, or agent-to-agent messages may spread throughout the graph and lead to unintended behavior, a breach of policy, or corruption of state. In order to mitigate these risks, this paper suggests a Cross-Agent Multimodal Provenanc- Aware Defense Framework whereby all the prompts, either user-generated or produced by upstream agents, are sanitized and all the outputs generated by an LLM are verified independently before being sent to downstream nodes. This framework contains a Text sanitizer agent, visual sanitizer agent, and output validator agent all coordinated by a provenance ledger, which keeps metadata of modality, source, and trust level throughout the entire agent network. This architecture makes sure that agent-to-agent communication abides by clear trust frames such such that injected instructions are not propagated down LangChain or GraphChain-style-workflows. The experimental assessments show that multimodal injection detection accuracy is significantly enhanced, and the cross-agent trust leakage is minimized, as well as, agentic execution pathways become stable. The framework, which expands the concept of provenance tracking and validation to the multi-agent orchestration, enhances the establishment of secure, understandable and reliable agentic AI systems.
Large language models (LLMs) are highly vulnerable to input confirmation bias. When a prompt implies a preferred answer, models often reinforce that bias rather than explore alternatives. This phenomenon remains underexplored, yet it is already harmful in base models and poses an even greater risk in multi-agent debate, where echo chambers reinforce bias instead of correction. We introduce Mixture of Latent Concept Experts (MoLaCE), a lightweight inference-time framework that addresses confirmation bias by mixing experts instantiated as different activation strengths over latent concepts that shape model responses. Our key insight is that, due to the compositional nature of language, differently phrased prompts reweight latent concepts in prompt-specific ways that affect factual correctness, so no single fixed intervention can be applied universally across inputs. This design enables a single LLM to emulate the benefits of debate internally while remaining computationally efficient and scalable. It can also be integrated into multi-agent debate frameworks to diversify perspectives and reduce correlated errors. We empirically show that it consistently reduces confirmation bias, improves robustness, and matches or surpasses multi-agent debate while requiring only a fraction of the computation.
Radio Access Network (RAN) slicing enables multiple logical networks to exist on top of the same physical infrastructure by allocating resources to distinct service groups, where radio resource scheduling plays a key role in ensuring compliance with slice-specific Service-Level Agreements (SLAs). Existing configuration-based or intent-driven Reinforcement Learning (RL) approaches usually rely on static mappings and SLA conversions. The current literature does not integrate natural language understanding with coordinated decision-making. To address these limitations, we propose an Agentic AI framework for 6G RAN slicing, driven by a super agent built using Hierarchical Decision Mamba (HDM) controllers and a Large Language Model (LLM). The super agent interprets operator intents and translates them into actionable goals using the LLM, which are used by HDM to coordinate inter-slice, intra-slice, and self-healing agents. Compared to transformer-based and reward-driven baselines, the proposed Agentic AI framework demonstrates consistent improvements across key performance indicators, including higher throughput, improved cell-edge performance, and reduced latency across different slices.
Most venture capital (VC) investments fail, while a few deliver outsized returns. Accurately predicting startup success requires synthesizing complex relational evidence, including company disclosures, investor track records, and investment network structures, through explicit reasoning to form coherent, interpretable investment theses. Traditional machine learning and graph neural networks both lack this reasoning capability. Large language models (LLMs) offer strong reasoning but face a modality mismatch with graphs. Recent graph-LLM methods target in-graph tasks where answers lie within the graph, whereas VC prediction is off-graph: the target exists outside the network. The core challenge is selecting graph paths that maximize predictor performance on an external objective while enabling step-by-step reasoning. We present MIRAGE-VC, a multi-perspective retrieval-augmented generation framework that addresses two obstacles: path explosion (thousands of candidate paths overwhelm LLM context) and heterogeneous evidence fusion (different startups need different analytical emphasis). Our information-gain-driven path retriever iteratively selects high-value neighbors, distilling investment networks into compact chains for explicit reasoning. A multi-agent architecture integrates three evidence streams via a learnable gating mechanism based on company attributes. Under strict anti-leakage controls, MIRAGE-VC achieves +5.0% F1 and +16.6% PrecisionAt5, and sheds light on other off-graph prediction tasks such as recommendation and risk assessment. Code: https://anonymous.4open.science/r/MIRAGE-VC-323F.
The software supply chain attacks are becoming more and more focused on trusted development and delivery procedures, so the conventional post-build integrity mechanisms cannot be used anymore. The available frameworks like SLSA, SBOM and in toto are majorly used to offer provenance and traceability but do not have the capabilities of actively identifying and removing vulnerabilities in software production. The current paper includes an example of agentic artificial intelligence (AI) based on autonomous software supply chain security that combines large language model (LLM)-based reasoning, reinforcement learning (RL), and multi-agent coordination. The suggested system utilizes specialized security agents coordinated with the help of LangChain and LangGraph, communicates with actual CI/CD environments with the Model Context Protocol (MCP), and documents all the observations and actions in a blockchain security ledger to ensure integrity and auditing. Reinforcement learning can be used to achieve adaptive mitigation strategies that consider the balance between security effectiveness and the operational overhead, and LLMs can be used to achieve semantic vulnerability analysis, as well as explainable decisions. This framework is tested based on simulated pipelines, as well as, actual world CI/CD integrations on GitHub Actions and Jenkins, including injection attacks, insecure deserialization, access control violations, and configuration errors. Experimental outcomes indicate better detection accuracy, shorter mitigation latency and reasonable build-time overhead than rule-based, provenance only and RL only baselines. These results show that agentic AI can facilitate the transition to self defending, proactive software supply chains rather than reactive verification ones.
Clinical diagnosis begins with doctor-patient interaction, during which physicians iteratively gather information, determine examination and refine differential diagnosis through patients' response. This dynamic clinical-reasoning process is poorly represented by existing LLM benchmarks that focus on static question-answering. To mitigate these gaps, recent methods explore dynamic medical frameworks involving interactive clinical dialogues. Although effective, they often rely on limited, contamination-prone datasets and lack granular, multi-level evaluation. In this work, we propose ClinDEF, a dynamic framework for assessing clinical reasoning in LLMs through simulated diagnostic dialogues. Grounded in a disease knowledge graph, our method dynamically generates patient cases and facilitates multi-turn interactions between an LLM-based doctor and an automated patient agent. Our evaluation protocol goes beyond diagnostic accuracy by incorporating fine-grained efficiency analysis and rubric-based assessment of diagnostic quality. Experiments show that ClinDEF effectively exposes critical clinical reasoning gaps in state-of-the-art LLMs, offering a more nuanced and clinically meaningful evaluation paradigm.
Modern AI models demand high-performance computation kernels. The growing complexity of LLMs, multimodal architectures, and recommendation systems, combined with techniques like sparsity and quantization, creates significant computational challenges. Moreover, frequent hardware updates and diverse chip architectures further complicate this landscape, requiring tailored kernel implementations for each platform. However, manual optimization cannot keep pace with these demands, creating a critical bottleneck in AI system development. Recent advances in LLM code generation capabilities have opened new possibilities for automating kernel development. In this work, we propose AKG kernel agent (AI-driven Kernel Generator), a multi-agent system that automates kernel generation, migration, and performance tuning. AKG kernel agent is designed to support multiple domain-specific languages (DSLs), including Triton, TileLang, CPP, and CUDA-C, enabling it to target different hardware backends while maintaining correctness and portability. The system's modular design allows rapid integration of new DSLs and hardware targets. When evaluated on KernelBench using Triton DSL across GPU and NPU backends, AKG kernel agent achieves an average speedup of 1.46$\times$ over PyTorch Eager baselines implementations, demonstrating its effectiveness in accelerating kernel development for modern AI workloads.
The flexibility and accuracy of methods for automatically counting objects in images and videos are limited by the way the object can be specified. While existing methods allow users to describe the target object with text and visual examples, the visual examples must be manually annotated inside the image, and there is no way to specify what not to count. To address these gaps, we introduce novel capabilities that expand how the target object can be specified. Specifically, we extend the prompt to enable what not to count to be described with text and/or visual examples, introduce the concept of `pseudo-exemplars' that automate the annotation of visual examples at inference, and extend counting models to accept visual examples from both natural and synthetic external images. We also use our new counting model, CountGD++, as a vision expert agent for an LLM. Together, these contributions expand the prompt flexibility of multi-modal open-world counting and lead to significant improvements in accuracy, efficiency, and generalization across multiple datasets. Code is available at https://github.com/niki-amini-naieni/CountGDPlusPlus.
Memory serves as the pivotal nexus bridging past and future, providing both humans and AI systems with invaluable concepts and experience to navigate complex tasks. Recent research on autonomous agents has increasingly focused on designing efficient memory workflows by drawing on cognitive neuroscience. However, constrained by interdisciplinary barriers, existing works struggle to assimilate the essence of human memory mechanisms. To bridge this gap, we systematically synthesizes interdisciplinary knowledge of memory, connecting insights from cognitive neuroscience with LLM-driven agents. Specifically, we first elucidate the definition and function of memory along a progressive trajectory from cognitive neuroscience through LLMs to agents. We then provide a comparative analysis of memory taxonomy, storage mechanisms, and the complete management lifecycle from both biological and artificial perspectives. Subsequently, we review the mainstream benchmarks for evaluating agent memory. Additionally, we explore memory security from dual perspectives of attack and defense. Finally, we envision future research directions, with a focus on multimodal memory systems and skill acquisition.
Large Language Model (LLM) agents, while proficient in the digital realm, face a significant gap in physical-world deployment due to the challenge of forming and maintaining a robust spatial mental model. We identify three core cognitive challenges hindering this transition: spatial reasoning, long-horizon state tracking via mental simulation, and active exploration under partial observation. To isolate and evaluate these faculties, we introduce CubeBench, a novel generative benchmark centered on the Rubik's Cube. CubeBench uses a three-tiered diagnostic framework that progressively assesses agent capabilities, from foundational state tracking with full symbolic information to active exploration with only partial visual data. Our experiments on leading LLMs reveal critical limitations, including a uniform 0.00% pass rate on all long-horizon tasks, exposing a fundamental failure in long-term planning. We also propose a diagnostic framework to isolate these cognitive bottlenecks by providing external solver tools. By analyzing the failure modes, we provide key insights to guide the development of more physically-grounded intelligent agents.
Multimodal Large Language Models (LLMs) introduce an emerging paradigm for medical imaging by interpreting scans through the lens of extensive clinical knowledge, offering a transformative approach to disease classification. This study presents a critical comparison between two fundamentally different AI architectures: the specialized open-source agent MedGemma and the proprietary large multimodal model GPT-4 for diagnosing six different diseases. The MedGemma-4b-it model, fine-tuned using Low-Rank Adaptation (LoRA), demonstrated superior diagnostic capability by achieving a mean test accuracy of 80.37% compared to 69.58% for the untuned GPT-4. Furthermore, MedGemma exhibited notably higher sensitivity in high-stakes clinical tasks, such as cancer and pneumonia detection. Quantitative analysis via confusion matrices and classification reports provides comprehensive insights into model performance across all categories. These results emphasize that domain-specific fine-tuning is essential for minimizing hallucinations in clinical implementation, positioning MedGemma as a sophisticated tool for complex, evidence-based medical reasoning.
Audiobook interpretations are attracting increasing attention, as they provide accessible and in-depth analyses of books that offer readers practical insights and intellectual inspiration. However, their manual creation process remains time-consuming and resource-intensive. To address this challenge, we propose AI4Reading, a multi-agent collaboration system leveraging large language models (LLMs) and speech synthesis technology to generate podcast, like audiobook interpretations. The system is designed to meet three key objectives: accurate content preservation, enhanced comprehensibility, and a logical narrative structure. To achieve these goals, we develop a framework composed of 11 specialized agents,including topic analysts, case analysts, editors, a narrator, and proofreaders that work in concert to explore themes, extract real world cases, refine content organization, and synthesize natural spoken language. By comparing expert interpretations with our system's output, the results show that although AI4Reading still has a gap in speech generation quality, the generated interpretative scripts are simpler and more accurate.
Semantic communications (SemCom), as one of the key technologies for 6G, is shifting networks from bit transmission to semantic information exchange. On this basis, introducing agentic artificial intelligence (AI) with perception, memory, reasoning, and action capabilities provides a practicable path to intelligent communications. This paper provides a systematic exposition of how agentic AI empowers SemCom from the perspectives of research foundations, system architecture, and application scenarios. We first provide a comprehensive review of existing studies by agent types, covering embedded agents, large language model (LLM)/large vision model (LVM) agents, and reinforcement learning (RL) agents. Additionally, we propose a unified agentic AI-enhanced SemCom framework covering the application layer, the semantic layer, and the cloud-edge collaboration layer, forming a closed loop from intent to encoding to transmission to decoding to action to evaluation. We also present several typical scenarios, including multi-vehicle collaborative perception, multi-robot cooperative rescue, and agentic operations for intellicise (intelligent and concise) networks. Furthermore, we introduce an agentic knowledge base (KB)-based joint source-channel coding case study, AKB-JSCC, where the source KB and channel KB are built by LLM/LVM agents and RL agents, respectively. Experimental results show that AKB-JSCC achieves higher information reconstruction quality under different channel conditions. Finally, we discuss future evolution and research directions, providing a reference for portable, verifiable, and controllable research and deployment of agentic SemCom.
A critical gap exists in LLM task-specific benchmarks. Thermal comfort, a sophisticated interplay of environmental factors and personal perceptions involving sensory integration and adaptive decision-making, serves as an ideal paradigm for evaluating real-world cognitive capabilities of AI systems. To address this, we propose TCEval, the first evaluation framework that assesses three core cognitive capacities of AI, cross-modal reasoning, causal association, and adaptive decision-making, by leveraging thermal comfort scenarios and large language model (LLM) agents. The methodology involves initializing LLM agents with virtual personality attributes, guiding them to generate clothing insulation selections and thermal comfort feedback, and validating outputs against the ASHRAE Global Database and Chinese Thermal Comfort Database. Experiments on four LLMs show that while agent feedback has limited exact alignment with humans, directional consistency improves significantly with a 1 PMV tolerance. Statistical tests reveal that LLM-generated PMV distributions diverge markedly from human data, and agents perform near-randomly in discrete thermal comfort classification. These results confirm the feasibility of TCEval as an ecologically valid Cognitive Turing Test for AI, demonstrating that current LLMs possess foundational cross-modal reasoning ability but lack precise causal understanding of the nonlinear relationships between variables in thermal comfort. TCEval complements traditional benchmarks, shifting AI evaluation focus from abstract task proficiency to embodied, context-aware perception and decision-making, offering valuable insights for advancing AI in human-centric applications like smart buildings.
Large Language Models (LLMs) often falter at complex planning tasks that require exploration and self-correction, as their linear reasoning process struggles to recover from early mistakes. While search algorithms like Monte Carlo Tree Search (MCTS) can explore alternatives, they are often ineffective when guided by sparse rewards and fail to leverage the rich semantic capabilities of LLMs. We introduce SPIRAL (Symbolic LLM Planning via Grounded and Reflective Search), a novel framework that embeds a cognitive architecture of three specialized LLM agents into an MCTS loop. SPIRAL's key contribution is its integrated planning pipeline where a Planner proposes creative next steps, a Simulator grounds the search by predicting realistic outcomes, and a Critic provides dense reward signals through reflection. This synergy transforms MCTS from a brute-force search into a guided, self-correcting reasoning process. On the DailyLifeAPIs and HuggingFace datasets, SPIRAL consistently outperforms the default Chain-of-Thought planning method and other state-of-the-art agents. More importantly, it substantially surpasses other state-of-the-art agents; for example, SPIRAL achieves 83.6% overall accuracy on DailyLifeAPIs, an improvement of over 16 percentage points against the next-best search framework, while also demonstrating superior token efficiency. Our work demonstrates that structuring LLM reasoning as a guided, reflective, and grounded search process yields more robust and efficient autonomous planners. The source code, full appendices, and all experimental data are available for reproducibility at the official project repository.
Machine learning (ML) underpins foundation models in finance, healthcare, and critical infrastructure, making them targets for data poisoning, model extraction, prompt injection, automated jailbreaking, and preference-guided black-box attacks that exploit model comparisons. Larger models can be more vulnerable to introspection-driven jailbreaks and cross-modal manipulation. Traditional cybersecurity lacks ML-specific threat modeling for foundation, multimodal, and RAG systems. Objective: Characterize ML security risks by identifying dominant TTPs, vulnerabilities, and targeted lifecycle stages. Methods: We extract 93 threats from MITRE ATLAS (26), AI Incident Database (12), and literature (55), and analyze 854 GitHub/Python repositories. A multi-agent RAG system (ChatGPT-4o, temp 0.4) mines 300+ articles to build an ontology-driven threat graph linking TTPs, vulnerabilities, and stages. Results: We identify unreported threats including commercial LLM API model stealing, parameter memorization leakage, and preference-guided text-only jailbreaks. Dominant TTPs include MASTERKEY-style jailbreaking, federated poisoning, diffusion backdoors, and preference optimization leakage, mainly impacting pre-training and inference. Graph analysis reveals dense vulnerability clusters in libraries with poor patch propagation. Conclusion: Adaptive, ML-specific security frameworks, combining dependency hygiene, threat intelligence, and monitoring, are essential to mitigate supply-chain and inference risks across the ML lifecycle.
Large language models are increasingly deployed in multi-agent workflows. We introduce Prompt Choreography, a framework that efficiently executes LLM workflows by maintaining a dynamic, global KV cache. Each LLM call can attend to an arbitrary, reordered subset of previously encoded messages. Parallel calls are supported. Though caching messages' encodings sometimes gives different results from re-encoding them in a new context, we show in diverse settings that fine-tuning the LLM to work with the cache can help it mimic the original results. Prompt Choreography significantly reduces per-message latency (2.0--6.2$\times$ faster time-to-first-token) and achieves substantial end-to-end speedups ($>$2.2$\times$) in some workflows dominated by redundant computation.
Density functional theory (DFT) and machine learning potentials (MLPs) are essential for predicting and understanding materials properties, yet preparing, executing, and analyzing these simulations typically requires extensive scripting, multi-step procedures, and significant high-performance computing (HPC) expertise. These challenges hinder reproducibility and slow down discovery. Here, we introduce Masgent, an AI-assisted materials simulation agent that unifies structure manipulation, automated VASP input generation, DFT workflow construction and analysis, fast MLP-based simulations, and lightweight machine learning (ML) utilities within a single platform. Powered by large language models (LLMs), Masgent enables researchers to perform complex simulation tasks through natural-language interaction, eliminating most manual scripting and reducing setup time from hours to seconds. By standardizing protocols and integrating advanced simulation and data-driven tools, Masgent democratizes access to state-of-the-art computational methodologies, accelerating hypothesis testing, pre-screening, and exploratory research for both new and experienced practitioners.
Large Language Models (LLMs) are increasingly used in complex knowledge work, yet linear transcript interfaces limit support for reflection. Schon's Reflective Practice distinguishes between reflection-in-action (during a task) and reflection-on-action (after a task), both benefiting from non-linear, revisitable representations of dialogue. ChatGraPhT is an interactive tool that shows dialogue as a visual map, allowing users to branch and merge ideas, edit past messages, and receive guidance that prompts deeper reflection. It supports non-linear, multi-path dialogue, while two agentic LLM assistants provide moment-to-moment and higher-level guidance. Our inquiry suggests that keeping the conversation structure visible, allowing branching and merging, and suggesting patterns or ways to combine ideas deepened user reflective engagement. Contributions are: (1) the design of a node-link, agentic LLM interface for reflective dialogue, and (2) transferable design knowledge on balancing structure and AI support to sustain reflection in complex, open-ended tasks.
The evolution of networked systems, driven by innovations in software-defined networking (SDN), network function virtualization (NFV), open radio access networks (O-RAN), and cloud-native architectures, is redefining both the operational landscape and the threat surface of critical infrastructures. This book offers an in-depth, interdisciplinary examination of how resilience must be re-conceptualized and re-engineered to address the multifaceted challenges posed by these transformations. Structured across six chapters, this book begins by surveying the contemporary risk landscape, identifying emerging cyber, physical, and AI-driven threats, and analyzing their implications for scalable, heterogeneous network environments. It then establishes rigorous definitions and evaluation frameworks for resilience, going beyond robustness and fault-tolerance to address adaptive, anticipatory, and retrospective mechanisms across diverse application domains. The core of the book delves into advanced paradigms and practical strategies for resilience, including zero trust architectures, game-theoretic threat modeling, and self-healing design principles. A significant portion is devoted to the role of artificial intelligence, especially reinforcement learning and large language models (LLMs), in enabling dynamic threat response, autonomous network control, and multi-agent coordination under uncertainty.
Large language model (LLM) agents have demonstrated strong capabilities in planning and tool use. Travel planning provides a natural and high-impact testbed for these capabilities, as it requires multi-step reasoning, iterative preference elicitation through interaction, and calls to external tools under evolving constraints. Prior work has studied LLMs on travel-planning tasks, but existing settings are limited in domain coverage and multi-turn interaction. As a result, they cannot support dynamic user-agent interaction and therefore fail to comprehensively assess agent capabilities. In this paper, we introduce TravelBench, a real-world travel-planning benchmark featuring multi-turn interaction and tool use. We collect user requests from real-world scenarios and construct three subsets-multi-turn, single-turn, and unsolvable-to evaluate different aspects of agent performance. For stable and reproducible evaluation, we build a controlled sandbox environment with 10 travel-domain tools, providing deterministic tool outputs for reliable reasoning. We evaluate multiple LLMs on TravelBench and conduct an analysis of their behaviors and performance. TravelBench offers a practical and reproducible benchmark for advancing LLM agents in travel planning.
Large Language Model (LLM) agents can increasingly automate complex reasoning through Test-Time Scaling (TTS), iterative refinement guided by reward signals. However, many real-world tasks involve multi-stage pipeline whose final outcomes lack verifiable rewards or sufficient data to train robust reward models, making judge-based refinement prone to accumulate error over stages. We propose Selective TTS, a process-based refinement framework that scales inference across different stages in multi-agent pipeline, instead of repeated refinement over time by prior work. By distributing compute across stages and pruning low-quality branches early using process-specific judges, Selective TTS mitigates the judge drift and stabilizes refinement. Grounded in the data science pipeline, we build an end-to-end multi-agent pipeline for generating visually insightful charts and report of given dataset, and design a reliable LLM-based judge model, aligned with human experts (Kendall's τ=0.55). Our proposed selective TTS then improves insight quality under a fixed compute budget, increasing mean scores from 61.64 to 65.86 while reducing variance. We hope our findings serve as the first step toward to scaling complex, open-ended tasks with unverifiable rewards, such as scientific discovery and story generation.
Due to the high value and high failure rate of startups, predicting their success has become a critical challenge across interdisciplinary research. Existing approaches typically model success prediction from the perspective of a single decision-maker, overlooking the collective dynamics of investor groups that dominate real-world venture capital (VC) decisions. In this paper, we propose SimVC-CAS, a novel collective agent system that simulates VC decision-making as a multi-agent interaction process. By designing role-playing agents and a GNN-based supervised interaction module, we reformulate startup financing prediction as a group decision-making task, capturing both enterprise fundamentals and the behavioral dynamics of potential investor networks. Each agent embodies an investor with unique traits and preferences, enabling heterogeneous evaluation and realistic information exchange through a graph-structured co-investment network. Using real-world data from PitchBook and under strict data leakage controls, we show that SimVC-CAS significantly improves predictive accuracy while providing interpretable, multiperspective reasoning, for example, approximately 25% relative improvement with respect to average precision@10. SimVC-CAS also sheds light on other complex group decision scenarios.
Agentic Reinforcement Learning (RL) enables Large Language Models (LLMs) to perform autonomous decision-making and long-term planning. Unlike standard LLM post-training, agentic RL workloads are highly heterogeneous, combining compute-intensive prefill phases, bandwidth-bound decoding, and stateful, CPU-heavy environment simulations. We argue that efficient agentic RL training requires disaggregated infrastructure to leverage specialized, best-fit hardware. However, naive disaggregation introduces substantial synchronization overhead and resource underutilization due to the complex dependencies between stages. We present RollArc, a distributed system designed to maximize throughput for multi-task agentic RL on disaggregated infrastructure. RollArc is built on three core principles: (1) hardware-affinity workload mapping, which routes compute-bound and bandwidth-bound tasks to bestfit GPU devices, (2) fine-grained asynchrony, which manages execution at the trajectory level to mitigate resource bubbles, and (3) statefulness-aware computation, which offloads stateless components (e.g., reward models) to serverless infrastructure for elastic scaling. Our results demonstrate that RollArc effectively improves training throughput and achieves 1.35-2.05\(\times\) end-to-end training time reduction compared to monolithic and synchronous baselines. We also evaluate RollArc by training a hundreds-of-billions-parameter MoE model for Qoder product on an Alibaba cluster with more than 3,000 GPUs, further demonstrating RollArc scalability and robustness. The code is available at https://github.com/alibaba/ROLL.
Large Language Models (LLMs) are increasingly deployed as automated tutors to address educator shortages; however, they often fail at pedagogical reasoning, frequently validating incorrect student solutions (sycophancy) or providing overly direct answers that hinder learning. We introduce Hierarchical Pedagogical Oversight (HPO), a framework that adapts structured adversarial synthesis to educational assessment. Unlike cooperative multi-agent systems that often drift toward superficial consensus, HPO enforces a dialectical separation of concerns: specialist agents first distill dialogue context, which then grounds a moderated, five-act debate between opposing pedagogical critics. We evaluate this framework on the MRBench dataset of 1,214 middle-school mathematics dialogues. Our 8B-parameter model achieves a Macro F1 of 0.845, outperforming GPT-4o (0.812) by 3.3% while using 20 times fewer parameters. These results establish adversarial reasoning as a critical mechanism for deploying reliable, low-compute pedagogical oversight in resource-constrained environments.
Analog circuit design remains a knowledge- and experience-intensive process that relies heavily on human intuition for topology generation and device parameter tuning. Existing LLM-based approaches typically depend on prompt-driven netlist generation or predefined topology templates, limiting their ability to satisfy complex specification requirements. We propose AnalogSAGE, an open-source self-evolving multi-agent framework that coordinates three-stage agent explorations through four stratified memory layers, enabling iterative refinement with simulation-grounded feedback. To support reproducibility and generality, we release the source code. Our benchmark spans ten specification-driven operational amplifier design problems of varying difficulty, enabling quantitative and cross-task comparison under identical conditions. Evaluated under the open-source SKY130 PDK with ngspice, AnalogSAGE achieves a 10$\times$ overall pass rate, a 48$\times$ Pass@1, and a 4$\times$ reduction in parameter search space compared with existing frameworks, demonstrating that stratified memory and grounded reasoning substantially enhance the reliability and autonomy of analog design automation in practice.
The proliferation of Large Language Models (LLMs) has catalyzed a shift towards autonomous agents capable of complex reasoning and tool use. However, current agent architectures are frequently constructed using imperative, ad hoc patterns. This results in brittle systems plagued by difficulties in state management, error handling, and concurrency. This paper introduces Monadic Context Engineering (MCE), a novel architectural paradigm leveraging the algebraic structures of Functors, Applicative Functors, and Monads to provide a formal foundation for agent design. MCE treats agent workflows as computational contexts where cross-cutting concerns, such as state propagation, short-circuiting error handling, and asynchronous execution, are managed intrinsically by the algebraic properties of the abstraction. We demonstrate how Monads enable robust sequential composition, how Applicatives provide a principled structure for parallel execution, and crucially, how Monad Transformers allow for the systematic composition of these capabilities. This layered approach enables developers to construct complex, resilient, and efficient AI agents from simple, independently verifiable components. We further extend this framework to describe Meta-Agents, which leverage MCE for generative orchestration, dynamically creating and managing sub-agent workflows through metaprogramming. Project Page: https://github.com/yifanzhang-pro/monadic-context-engineering.
With the significant increase in enrollment in computing-related programs over the past 20 years, lecture sizes have grown correspondingly. In large lectures, instructors face challenges on identifying students' knowledge gaps timely, which is critical for effective teaching. Existing classroom response systems rely on instructor-initiated interactions, which limits their ability to capture the spontaneous knowledge gaps that naturally emerge during lectures. With the widespread adoption of LLMs among students, we recognize these student-AI dialogues as a valuable, student-centered data source for identifying knowledge gaps. In this idea paper, we propose QueryQuilt, a multi-agent LLM framework that automatically detects common knowledge gaps in large-scale lectures by analyzing students' chat logs with AI assistants. QueryQuilt consists of two key components: (1) a Dialogue Agent that responds to student questions while employing probing questions to reveal underlying knowledge gaps, and (2) a Knowledge Gap Identification Agent that systematically analyzes these dialogues to identify knowledge gaps across the student population. By generating frequency distributions of identified gaps, instructors can gain comprehensive insights into class-wide understanding. Our evaluation demonstrates promising results, with QueryQuilt achieving 100% accuracy in identifying knowledge gaps among simulated students and 95% completeness when tested on real student-AI dialogue data. These initial findings indicate the system's potential for facilitate teaching in authentic learning environments. We plan to deploy QueryQuilt in actual classroom settings for comprehensive evaluation, measuring its detection accuracy and impact on instruction.