Code translation is a crucial task in software development and maintenance. While recent advancements in large language models (LLMs) have improved automated code translation accuracy, these gains often come at the cost of increased inference latency, hindering real-world development workflows that involve human-in-the-loop inspection. To address this trade-off, we propose EffiReasonTrans, a training framework designed to improve translation accuracy while balancing inference latency. We first construct a high-quality reasoning-augmented dataset by prompting a stronger language model, DeepSeek-R1, to generate intermediate reasoning and target translations. Each (source code, reasoning, target code) triplet undergoes automated syntax and functionality checks to ensure reliability. Based on this dataset, we employ a two-stage training strategy: supervised fine-tuning on reasoning-augmented samples, followed by reinforcement learning to further enhance accuracy and balance inference latency. We evaluate EffiReasonTrans on six translation pairs. Experimental results show that it consistently improves translation accuracy (up to +49.2% CA and +27.8% CodeBLEU compared to the base model) while reducing the number of generated tokens (up to -19.3%) and lowering inference latency in most cases (up to -29.0%). Ablation studies further confirm the complementary benefits of the two-stage training framework. Additionally, EffiReasonTrans demonstrates improved translation accuracy when integrated into agent-based frameworks. Our code and data are available at https://github.com/DeepSoftwareAnalytics/EffiReasonTrans.
Traditional machine learning has advanced polymer discovery, yet direct generation of chemically valid and synthesizable polymers without exhaustive enumeration remains a challenge. Here we present polyT5, an encoder-decoder chemical language model based on the T5 architecture, trained to understand and generate polymer structures. polyT5 enables both property prediction and the targeted generation of polymers conditioned on desired property values. We demonstrate its utility for dielectric polymer design, seeking candidates with dielectric constant >3, bandgap >4 eV, and glass transition temperature >400 K, alongside melt-processability and solubility requirements. From over 20,000 generated promising candidates, one was experimentally synthesized and validated, showing strong agreement with predictions. To further enhance usability, we integrated polyT5 within an agentic AI framework that couples it with a general-purpose LLM, allowing natural language interaction for property prediction and generative design. Together, these advances establish a versatile and accessible framework for accelerated polymer discovery.
Reinforcement learning with verifiable rewards (RLVR) has become the mainstream technique for training LLM agents. However, RLVR highly depends on well-crafted task queries and corresponding ground-truth answers to provide accurate rewards, which requires massive human efforts and hinders the RL scaling processes, especially under agentic scenarios. Although a few recent works explore task synthesis methods, the difficulty of generated agentic tasks can hardly be controlled to provide effective RL training advantages. To achieve agentic RLVR with higher scalability, we explore self-play training for deep search agents, in which the learning LLM utilizes multi-turn search engine calling and acts simultaneously as both a task proposer and a problem solver. The task proposer aims to generate deep search queries with well-defined ground-truth answers and increasing task difficulty. The problem solver tries to handle the generated search queries and output the correct answer predictions. To ensure that each generated search query has accurate ground truth, we collect all the searching results from the proposer's trajectory as external knowledge, then conduct retrieval-augmentation generation (RAG) to test whether the proposed query can be correctly answered with all necessary search documents provided. In this search self-play (SSP) game, the proposer and the solver co-evolve their agent capabilities through both competition and cooperation. With substantial experimental results, we find that SSP can significantly improve search agents' performance uniformly on various benchmarks without any supervision under both from-scratch and continuous RL training setups. The code is at https://github.com/Alibaba-Quark/SSP.
Recent advances in large language models (LLMs) have enabled progress in agentic coding, where models autonomously reason, plan, and act within interactive software development workflows. However, bridging the gap between static text-based training and dynamic real-world agentic execution remains a core challenge. In this technical report, we present KAT-Coder, a large-scale agentic code model trained through a multi-stage curriculum encompassing Mid-Term Training, Supervised Fine-Tuning (SFT), Reinforcement Fine-Tuning (RFT), and Reinforcement-to-Deployment Adaptation. The Mid-Term stage enhances reasoning, planning, and reflection capabilities through a corpus of real software engineering data and synthetic agentic interactions. The SFT stage constructs a million-sample dataset balancing twenty programming languages, ten development contexts, and ten task archetypes. The RFT stage introduces a novel multi-ground-truth reward formulation for stable and sample-efficient policy optimization. Finally, the Reinforcement-to-Deployment phase adapts the model to production-grade IDE environments using Error-Masked SFT and Tree-Structured Trajectory Training. In summary, these stages enable KAT-Coder to achieve robust tool-use reliability, instruction alignment, and long-context reasoning, forming a deployable foundation for real-world intelligent coding agents. Our KAT series 32B model, KAT-Dev, has been open-sourced on https://huggingface.co/Kwaipilot/KAT-Dev.
Recent surges in LLM-driven intelligent systems largely overlook decades of foundational multi-agent systems (MAS) research, resulting in frameworks with critical limitations such as centralization and inadequate trust and communication protocols. This paper introduces the Fetch.ai architecture, an industrial-strength platform designed to bridge this gap by facilitating the integration of classical MAS principles with modern AI capabilities. We present a novel, multi-layered solution built on a decentralized foundation of on-chain blockchain services for verifiable identity, discovery, and transactions. This is complemented by a comprehensive development framework for creating secure, interoperable agents, a cloud-based platform for deployment, and an intelligent orchestration layer where an agent-native LLM translates high-level human goals into complex, multi-agent workflows. We demonstrate the deployed nature of this system through a decentralized logistics use case where autonomous agents dynamically discover, negotiate, and transact with one another securely. Ultimately, the Fetch.ai stack provides a principled architecture for moving beyond current agent implementations towards open, collaborative, and economically sustainable multi-agent ecosystems.
Large Language Models (LLMs) are increasingly deployed in complex multi-agent applications that use external function calls. This workload creates severe performance challenges for the KV Cache: space contention leads to the eviction of critical agents' caches and time underutilization leaves the cache of agents stalled on long-running tool calls idling in GPU memory. We present Tokencake, a KV-Cache-centric serving framework that co-optimizes scheduling and memory management with an agent-aware design. Tokencake's Space Scheduler uses dynamic memory partitioning to shield critical agents from contention, while its Time Scheduler employs a proactive offload and predictive upload mechanism to repurpose GPU memory during function call stalls. Our evaluation on representative multi-agent benchmarks shows that Tokencake can reduce end-to-end latency by over 47.06%, improve effective GPU memory utilization by up to 16.9% compared to vLLM.
Phishing websites remain a significant cybersecurity threat, necessitating accurate and cost-effective detection mechanisms. In this paper, we present CLASP, a novel system that effectively identifies phishing websites by leveraging multiple intelligent agents, built using large language models (LLMs), to analyze different aspects of a web resource. The system processes URLs or QR codes, employing specialized LLM-based agents that evaluate the URL structure, webpage screenshot, and HTML content to predict potential phishing threats. To optimize performance while minimizing operational costs, we experimented with multiple combination strategies for agent-based analysis, ultimately designing a strategic combination that ensures the per-website evaluation expense remains minimal without compromising detection accuracy. We tested various LLMs, including Gemini 1.5 Flash and GPT-4o mini, to build these agents and found that Gemini 1.5 Flash achieved the best performance with an F1 score of 83.01% on a newly curated dataset. Also, the system maintained an average processing time of 2.78 seconds per website and an API cost of around $3.18 per 1,000 websites. Moreover, CLASP surpasses leading previous solutions, achieving over 40% higher recall and a 20% improvement in F1 score for phishing detection on the collected dataset. To support further research, we have made our dataset publicly available, supporting the development of more advanced phishing detection systems.
Automating quantitative trading strategy development in dynamic markets is challenging, especially with increasing demand for personalized investment solutions. Existing methods often fail to explore the vast strategy space while preserving the diversity essential for robust performance across changing market conditions. We present QuantEvolve, an evolutionary framework that combines quality-diversity optimization with hypothesis-driven strategy generation. QuantEvolve employs a feature map aligned with investor preferences, such as strategy type, risk profile, turnover, and return characteristics, to maintain a diverse set of effective strategies. It also integrates a hypothesis-driven multi-agent system to systematically explore the strategy space through iterative generation and evaluation. This approach produces diverse, sophisticated strategies that adapt to both market regime shifts and individual investment needs. Empirical results show that QuantEvolve outperforms conventional baselines, validating its effectiveness. We release a dataset of evolved strategies to support future research.
Multi-agent systems powered by large language models are advancing rapidly, yet the tension between mutual trust and security remains underexplored. We introduce and empirically validate the Trust-Vulnerability Paradox (TVP): increasing inter-agent trust to enhance coordination simultaneously expands risks of over-exposure and over-authorization. To investigate this paradox, we construct a scenario-game dataset spanning 3 macro scenes and 19 sub-scenes, and run extensive closed-loop interactions with trust explicitly parameterized. Using Minimum Necessary Information (MNI) as the safety baseline, we propose two unified metrics: Over-Exposure Rate (OER) to detect boundary violations, and Authorization Drift (AD) to capture sensitivity to trust levels. Results across multiple model backends and orchestration frameworks reveal consistent trends: higher trust improves task success but also heightens exposure risks, with heterogeneous trust-to-risk mappings across systems. We further examine defenses such as Sensitive Information Repartitioning and Guardian-Agent enablement, both of which reduce OER and attenuate AD. Overall, this study formalizes TVP, establishes reproducible baselines with unified metrics, and demonstrates that trust must be modeled and scheduled as a first-class security variable in multi-agent system design.
The paradigm of LLM-as-a-judge is emerging as a scalable and efficient alternative to human evaluation, demonstrating strong performance on well-defined tasks. However, its reliability in open-ended tasks with dynamic environments and complex interactions remains unexplored. To bridge the gap, we introduce WebDevJudge, a systematic benchmark for assessing LLM-as-a-judge performance in web development, with support for both non-interactive evaluation based on static observations and continuous interactive evaluation with a dynamic web environment. WebDevJudge comprises human preference labels over paired web implementations, annotated with structured and query-grounded rubrics to ensure high-quality ground truth. Using this benchmark, we comprehensively evaluate various evaluators, including LLMs, MLLMs, and agentic workflows. We systematically investigate the impact of different paradigms and guidance mechanisms. Our experiments reveal a significant gap between LLM judges and human experts. In-depth analysis indicates this gap stems from fundamental model limitations, including failures in recognizing functional equivalence, verifying task feasibility, and mitigating bias. Overall, WebDevJudge presents a significant challenge to LLM-as-a-judge, offering insights to guide future research toward developing more reliable and capable automated evaluators for complicated scenarios. Code and data are available at https://github.com/lcy2723/WebDevJudge.
In this paper, we present SOCIA-Nabla, an end-to-end, agentic framework that treats simulator construction asinstance optimization over code within a textual computation graph. Specialized LLM-driven agents are embedded as graph nodes, and a workflow manager executes a loss-driven loop: code synthesis -> execution -> evaluation -> code repair. The optimizer performs Textual-Gradient Descent (TGD), while human-in-the-loop interaction is reserved for task-spec confirmation, minimizing expert effort and keeping the code itself as the trainable object. Across three CPS tasks, i.e., User Modeling, Mask Adoption, and Personal Mobility, SOCIA-Nabla attains state-of-the-art overall accuracy. By unifying multi-agent orchestration with a loss-aligned optimization view, SOCIA-Nabla converts brittle prompt pipelines into reproducible, constraint-aware simulator code generation that scales across domains and simulation granularities. This work is under review, and we will release the code soon.
Large Language Models (LLMs) increasingly rely on emerging protocols such as the Model Context Protocol (MCP) to invoke external tools and services. However, current tool routing mechanisms remain fragile because they only consider functional matching between users' queries and tools. In practice, user intent expressed through queries can be vague or underspecified, and the actual Quality of Experience (QoE) also depends on external factors such as link latency and server availability that are not captured by semantics alone. To address this challenge, we propose JAUNT, a framework for Joint Alignment of User intent and Network state in QoE-centric Tool routing. JAUNT introduces a dual-view alignment strategy that interprets user intent while employing LLM agents to construct network profiles, mapping numerical performance indicators into the semantic space to guide routing. We further design a benchmark that integrates diverse user request patterns with heterogeneous network states, enabling systematic evaluation of QoE outcomes. Experimental results show that JAUNT significantly improves QoE compared with several baselines, demonstrating the importance of aligning both intent and network state for scalable LLM service orchestration.
Object-goal navigation (ObjNav) tasks an agent with navigating to the location of a specific object in an unseen environment. Embodied agents equipped with large language models (LLMs) and online constructed navigation maps can perform ObjNav in a zero-shot manner. However, existing agents heavily rely on giant LLMs on the cloud, e.g., GPT-4, while directly switching to small LLMs, e.g., LLaMA3.2-11b, suffer from significant success rate drops due to limited model capacity for understanding complex navigation maps, which prevents deploying ObjNav on local devices. At the same time, the long prompt introduced by the navigation map description will cause high planning latency on local devices. In this paper, we propose EfficientNav to enable on-device efficient LLM-based zero-shot ObjNav. To help the smaller LLMs better understand the environment, we propose semantics-aware memory retrieval to prune redundant information in navigation maps. To reduce planning latency, we propose discrete memory caching and attention-based memory clustering to efficiently save and re-use the KV cache. Extensive experimental results demonstrate that EfficientNav achieves 11.1% improvement in success rate on HM3D benchmark over GPT-4-based baselines, and demonstrates 6.7x real-time latency reduction and 4.7x end-to-end latency reduction over GPT-4 planner. Our code will be released soon.
In this study, we present a solution for the modernization of legacy applications, an area of code generation where LLM-based multi-agent systems are proving essential for complex multi-phased tasks. Legacy applications often contain deprecated components that create compatibility, security, and reliability risks, but high resource costs make companies hesitate to update. We take a step forward to integrate an LLM-based multi-agent system as part of a legacy web application update to provide a cost-effective solution to update legacy applications autonomously. We propose a multi-agent system named a Verifying Agent Pipeline Updater (VAPU), which is designed to update code files in phases while simulating different roles in a software development team. In our previous study, we evaluated the system for legacy version updates by using six legacy web application view files by resulting errors and accomplished requirements. This study extends the previous evaluation of a multi-agent pipeline system by extending the evaluation of VAPU from a single LLM to five LLMs and using the temperature parameter in both 0 to 1 settings. Additionally, we tested the system with 20 open-source Python GitHub projects. The results of the evaluation were compared to Zero-Shot Learning (ZSL) and One-Shot Learning (OSL) prompts. The extended evaluation of VAPU showed that particularly in a low-temperature VAPU can get similar level of error count compared to the ZSL/OSL prompts but with a higher level of fulfilled requirements, depending on the LLM. VAPU showed up to 22.5% increase in the succeeding Python file update requirements compared to ZSL/OSL prompts. The study indicates that an LLM-based multi-agent system is a capable solution to update components of a legacy application autonomously.
Control algorithms in production environments typically require domain experts to tune their parameters and logic for specific scenarios. However, existing research predominantly focuses on algorithmic performance under ideal or default configurations, overlooking the critical aspect of Tuning Potential. To bridge this gap, we introduce Crucible, an agent that employs an LLM-driven, multi-level expert simulation to turn algorithms and defines a formalized metric to quantitatively evaluate their Tuning Potential. We demonstrate Crucible's effectiveness across a wide spectrum of case studies, from classic control tasks to complex computer systems, and validate its findings in a real-world deployment. Our experimental results reveal that Crucible systematically quantifies the tunable space across different algorithms. Furthermore, Crucible provides a new dimension for algorithm analysis and design, which ultimately leads to performance improvements. Our code is available at https://github.com/thu-media/Crucible.
Large Language Models (LLMs) have shown great promise in automating data analytics tasks by interpreting natural language queries and generating multi-operation execution plans. However, existing LLM-agent-based analytics frameworks operate under the assumption of centralized data access, offering little to no privacy protection. In contrast, federated analytics (FA) enables privacy-preserving computation across distributed data sources, but lacks support for natural language input and requires structured, machine-readable queries. In this work, we present LAFA, the first system that integrates LLM-agent-based data analytics with FA. LAFA introduces a hierarchical multi-agent architecture that accepts natural language queries and transforms them into optimized, executable FA workflows. A coarse-grained planner first decomposes complex queries into sub-queries, while a fine-grained planner maps each subquery into a Directed Acyclic Graph of FA operations using prior structural knowledge. To improve execution efficiency, an optimizer agent rewrites and merges multiple DAGs, eliminating redundant operations and minimizing computational and communicational overhead. Our experiments demonstrate that LAFA consistently outperforms baseline prompting strategies by achieving higher execution plan success rates and reducing resource-intensive FA operations by a substantial margin. This work establishes a practical foundation for privacy-preserving, LLM-driven analytics that supports natural language input in the FA setting.
We present a probabilistic intent modeling framework for large language model (LLM) agents in multi-turn social dialogue. The framework maintains a belief distribution over a partner's latent intentions, initialized from contextual priors and dynamically updated through likelihood estimation after each utterance. The evolving distribution provides additional contextual grounding for the policy, enabling adaptive dialogue strategies under uncertainty. Preliminary experiments in the SOTOPIA environment show consistent improvements: the proposed framework increases the Overall score by 9.0% on SOTOPIA-All and 4.1% on SOTOPIA-Hard compared with the Qwen2.5-7B baseline, and slightly surpasses an oracle agent that directly observes partner intentions. These early results suggest that probabilistic intent modeling can contribute to the development of socially intelligent LLM agents.
Chain-of-Thought (CoT) is widely applied to improve the LLM capability in math, coding and reasoning tasks. However, its performance is limited for open-domain tasks since there are no clearly defined reasoning steps or logical transitions. To mitigate such challenges, we propose another prompt-based paradigm called Chain of Conceptual Thought (CoCT), where the LLM first tags a concept, then generates the detailed content. The chain of concepts is allowed within the utterance, encouraging the LLM's deep and strategic thinking. We experiment with this paradigm in daily and emotional support conversations where the concept is comprised of emotions, strategies and topics. Automatic, human and model evaluations suggest that CoCT surpasses baselines such as Self-Refine, ECoT, ToT, SoT and RAG, suggesting a potential effective prompt-based paradigm of LLM for a wider scope of tasks.
This paper proposes Memory-Augmented State Machine Prompting (MASMP), a novel framework for LLM agents in real-time strategy games. Addressing key challenges like hallucinations and fragmented decision-making in existing approaches, MASMP integrates state machine prompting with memory mechanisms to unify structured actions with long-term tactical coherence. The framework features: (1) a natural language-driven state machine architecture that guides LLMs to emulate finite state machines and behavior trees through prompts, and (2) a lightweight memory module preserving strategic variables (e.g., tactics, priority units) across decision cycles. Experiments in StarCraft II demonstrate MASMP's 60% win rate against the hardest built-in AI (Lv7), vastly outperforming baselines (0%). Case studies reveal the method retains LLMs' semantic comprehension while resolving the "Knowing-Doing Gap" through strict state-action mapping, achieving both interpretability and FSM-like reliability. This work establishes a new paradigm for combining neural and symbolic AI in complex decision-making.
Large Language Models (LLMs) frequently generate buggy code with complex logic errors that are challenging to diagnose. While existing LLM-based self-repair approaches conduct intensive static semantic analysis or reply on superficial execution logs, they miss the in-depth runtime behaviors that often expose bug root causes-lacking the interactive dynamic analysis capabilities that make human debugging effective. We present InspectCoder, the first agentic program repair system that empowers LLMs to actively conduct dynamic analysis via interactive debugger control. Our dual-agent framework enables strategic breakpoint placement, targeted state inspection, and incremental runtime experimentation within stateful debugger sessions. Unlike existing methods that follow fixed log collection procedures, InspectCoder adaptively inspects and perturbs relevant intermediate states at runtime, and leverages immediate process rewards from debugger feedback to guide multi-step reasoning, transforming LLM debugging paradigm from blind trial-and-error into systematic root cause diagnosis. We conduct comprehensive experiments on two challenging self-repair benchmarks: BigCodeBench-R and LiveCodeBench-R. InspectCoder achieves 5.10%-60.37% relative improvements in repair accuracy over the strongest baseline, while delivering 1.67x-2.24x superior bug-fix efficiency respectively. We also contribute InspectWare, an open-source middleware that abstracts debugger complexities and maintains stateful debugging sessions across mainstream Python testing frameworks. Our work provides actionable insight into the interactive LLM-debugger systems, demonstrating the significant potential of LLM-driven dynamic analysis for automated software engineering.
As large language model (LLM) agents increasingly automate complex web tasks, they boost productivity while simultaneously introducing new security risks. However, relevant studies on web agent attacks remain limited. Existing red-teaming approaches mainly rely on manually crafted attack strategies or static models trained offline. Such methods fail to capture the underlying behavioral patterns of web agents, making it difficult to generalize across diverse environments. In web agent attacks, success requires the continuous discovery and evolution of attack strategies. To this end, we propose Genesis, a novel agentic framework composed of three modules: Attacker, Scorer, and Strategist. The Attacker generates adversarial injections by integrating the genetic algorithm with a hybrid strategy representation. The Scorer evaluates the target web agent's responses to provide feedback. The Strategist dynamically uncovers effective strategies from interaction logs and compiles them into a continuously growing strategy library, which is then re-deployed to enhance the Attacker's effectiveness. Extensive experiments across various web tasks show that our framework discovers novel strategies and consistently outperforms existing attack baselines.
Incentivizing the reasoning ability of Multimodal Large Language Models (MLLMs) is essential for medical applications to transparently analyze medical scans and provide reliable diagnosis. However, existing medical MLLMs rely solely on internal knowledge during reasoning, leading to hallucinated reasoning and factual inaccuracies when encountering cases beyond their training scope. Although recent Agentic Retrieval-Augmented Generation (RAG) methods elicit the medical model's proactive retrieval ability during reasoning, they are confined to unimodal LLMs, neglecting the crucial visual information during reasoning and retrieval. Consequently, we propose the first Multimodal Medical Reasoning-with-Retrieval framework, Med-RwR, which actively retrieves external knowledge by querying observed symptoms or domain-specific medical concepts during reasoning. Specifically, we design a two-stage reinforcement learning strategy with tailored rewards that stimulate the model to leverage both visual diagnostic findings and textual clinical information for effective retrieval. Building on this foundation, we further propose a Confidence-Driven Image Re-retrieval (CDIR) method for test-time scaling when low prediction confidence is detected. Evaluation on various public medical benchmarks demonstrates Med-RwR's significant improvements over baseline models, proving the effectiveness of enhancing reasoning capabilities with external knowledge integration. Furthermore, Med-RwR demonstrates remarkable generalizability to unfamiliar domains, evidenced by 8.8% performance gain on our proposed EchoCardiography Benchmark (ECBench), despite the scarcity of echocardiography data in the training corpus. Our data, model, and codes will be made publicly available at https://github.com/xmed-lab/Med-RwR.
Food insecurity remains a persistent public health emergency in the United States, tightly interwoven with chronic disease, mental illness, and opioid misuse. Yet despite the existence of thousands of food banks and pantries, access remains fragmented: 1) current retrieval systems depend on static directories or generic search engines, which provide incomplete and geographically irrelevant results; 2) LLM-based chatbots offer only vague nutritional suggestions and fail to adapt to real-world constraints such as time, mobility, and transportation; and 3) existing food recommendation systems optimize for culinary diversity but overlook survival-critical needs of food-insecure populations, including immediate proximity, verified availability, and contextual barriers. These limitations risk leaving the most vulnerable individuals, those experiencing homelessness, addiction, or digital illiteracy, unable to access urgently needed resources. To address this, we introduce Food4All, the first multi-agent framework explicitly designed for real-time, context-aware free food retrieval. Food4All unifies three innovations: 1) heterogeneous data aggregation across official databases, community platforms, and social media to provide a continuously updated pool of food resources; 2) a lightweight reinforcement learning algorithm trained on curated cases to optimize for both geographic accessibility and nutritional correctness; and 3) an online feedback loop that dynamically adapts retrieval policies to evolving user needs. By bridging information acquisition, semantic analysis, and decision support, Food4All delivers nutritionally annotated and guidance at the point of need. This framework establishes an urgent step toward scalable, equitable, and intelligent systems that directly support populations facing food insecurity and its compounding health risks.
Test suites in real-world projects are often large and achieve high code coverage, yet they remain insufficient for detecting all bugs. The abundance of unresolved issues in open-source project trackers highlights this gap. While regression tests are typically designed to ensure past functionality is preserved in the new version, they can also serve a complementary purpose: debugging the current version. Specifically, regression tests can (1) enhance the generation of reproduction tests for newly reported issues, and (2) validate that patches do not regress existing functionality. We present TestPrune, a fully automated technique that leverages issue tracker reports and strategically reuses regression tests for both bug reproduction and patch validation. A key contribution of TestPrune is its ability to automatically minimize the regression suite to a small, highly relevant subset of tests. Due to the predominance of LLM-based debugging techniques, this minimization is essential as large test suites exceed context limits, introduce noise, and inflate inference costs. TestPrune can be plugged into any agentic bug repair pipeline and orthogonally improve overall performance. As a proof of concept, we show that TestPrune leads to a 6.2%-9.0% relative increase in issue reproduction rate within the Otter framework and a 9.4% - 12.9% relative increase in issue resolution rate within the Agentless framework on SWE-Bench Lite and SWE-Bench Verified benchmarks, capturing fixes that were correctly produced by agents but not submitted as final patches. Compared to the benefits, the cost overhead of using TestPrune is minimal, i.e., \$0.02 and \$0.05 per SWE-Bench instance, using GPT-4o and Claude-3.7-Sonnet models, respectively.
Inference-time computation is a critical yet challenging paradigm for enhancing the reasoning performance of large language models (LLMs). While existing strategies improve reasoning stability and consistency, they suffer from notable limitations: self-correction often reinforces the model's initial biases, and Multi-Agent Collaboration (MAC) often fails due to the lack of efficient coordination mechanisms, leading to collective errors. Although high-performing verifiers can detect reasoning errors, making them reliable requires substantial training. To address these challenges, we introduce a novel inference-time framework, Adaptive Coopetition (AdCo), in which LLM agents utilize an adaptive, UCB-based "coopetition" mechanism. At each round, agents leverage coarse verifier signals to determine whether to collaborate or compete, and iteratively refine their reasoning based on peer feedback. Without relying on high-performance verifiers, our adaptive strategy achieves significant performance gains on mathematical reasoning benchmarks, yielding a 20% relative improvement over baselines on the more challenging dataset. Our approach remains robust and consistent in terms of accuracy under different sample sizes and configurations. This adaptive, signal-guided "coopetition" framework enhances reasoning robustness by leveraging both model knowledge diversity and reasoning trace measures, while also promoting uncertainty-driven exploration, especially when participants have comparable capabilities. From this perspective, our work offers a fresh lens on inference-time computation and paves the way for more resilient multi-agent LLM systems. Our code is available at: https://github.com/AdCo-Research/adaptive-coopetition.
Simulating consumer decision-making is vital for designing and evaluating marketing strategies before costly real-world deployment. However, post-event analyses and rule-based agent-based models (ABMs) struggle to capture the complexity of human behavior and social interaction. We introduce an LLM-powered multi-agent simulation framework that models consumer decisions and social dynamics. Building on recent advances in large language model simulation in a sandbox environment, our framework enables generative agents to interact, express internal reasoning, form habits, and make purchasing decisions without predefined rules. In a price-discount marketing scenario, the system delivers actionable strategy-testing outcomes and reveals emergent social patterns beyond the reach of conventional methods. This approach offers marketers a scalable, low-risk tool for pre-implementation testing, reducing reliance on time-intensive post-event evaluations and lowering the risk of underperforming campaigns.
Small Language Models (SLMs) offer compelling advantages in deployment cost and latency, but their accuracy often lags behind larger models, particularly for complex domain-specific tasks. While supervised fine-tuning can help bridge this performance gap, it requires substantial manual effort in data preparation and iterative optimization. We present PaDA-Agent (Pattern-guided Data Augmentation Agent), an evaluation-driven approach that streamlines the data augmentation process for SLMs through coordinated operations. Unlike state-of-the-art approaches that focus on model training errors only and generating error-correcting samples, PaDA-Agent discovers failure patterns from the validation data via evaluations and drafts targeted data augmentation strategies aiming to directly reduce the generalization gap. Our experimental results demonstrate significant improvements over state-of-the-art LLM-based data augmentation approaches for Llama 3.2 1B Instruct model fine-tuning.
As large language models (LLMs) are increasingly used for code generation, concerns over the security risks have grown substantially. Early research has primarily focused on red teaming, which aims to uncover and evaluate vulnerabilities and risks of CodeGen models. However, progress on the blue teaming side remains limited, as developing defense requires effective semantic understanding to differentiate the unsafe from the safe. To fill in this gap, we propose BlueCodeAgent, an end-to-end blue teaming agent enabled by automated red teaming. Our framework integrates both sides: red teaming generates diverse risky instances, while the blue teaming agent leverages these to detect previously seen and unseen risk scenarios through constitution and code analysis with agentic integration for multi-level defense. Our evaluation across three representative code-related tasks--bias instruction detection, malicious instruction detection, and vulnerable code detection--shows that BlueCodeAgent achieves significant gains over the base models and safety prompt-based defenses. In particular, for vulnerable code detection tasks, BlueCodeAgent integrates dynamic analysis to effectively reduce false positives, a challenging problem as base models tend to be over-conservative, misclassifying safe code as unsafe. Overall, BlueCodeAgent achieves an average 12.7\% F1 score improvement across four datasets in three tasks, attributed to its ability to summarize actionable constitutions that enhance context-aware risk detection. We demonstrate that the red teaming benefits the blue teaming by continuously identifying new vulnerabilities to enhance defense performance.
As users increasingly turn to large language model (LLM) based web agents to automate online tasks, agents may encounter dark patterns: deceptive user interface designs that manipulate users into making unintended decisions. Although dark patterns primarily target human users, their potentially harmful impacts on LLM-based generalist web agents remain unexplored. In this paper, we present the first study that investigates the impact of dark patterns on the decision-making process of LLM-based generalist web agents. To achieve this, we introduce LiteAgent, a lightweight framework that automatically prompts agents to execute tasks while capturing comprehensive logs and screen-recordings of their interactions. We also present TrickyArena, a controlled environment comprising web applications from domains such as e-commerce, streaming services, and news platforms, each containing diverse and realistic dark patterns that can be selectively enabled or disabled. Using LiteAgent and TrickyArena, we conduct multiple experiments to assess the impact of both individual and combined dark patterns on web agent behavior. We evaluate six popular LLM-based generalist web agents across three LLMs and discover that when there is a single dark pattern present, agents are susceptible to it an average of 41% of the time. We also find that modifying dark pattern UI attributes through visual design changes or HTML code adjustments and introducing multiple dark patterns simultaneously can influence agent susceptibility. This study emphasizes the need for holistic defense mechanisms in web agents, encompassing both agent-specific protections and broader web safety measures.
This paper explores the application of Large Language Models (LLMs) and reasoning to predict Dungeons & Dragons (DnD) player actions and format them as Avrae Discord bot commands. Using the FIREBALL dataset, we evaluated a reasoning model, DeepSeek-R1-Distill-LLaMA-8B, and an instruct model, LLaMA-3.1-8B-Instruct, for command generation. Our findings highlight the importance of providing specific instructions to models, that even single sentence changes in prompts can greatly affect the output of models, and that instruct models are sufficient for this task compared to reasoning models.