This work examines whether activating latent subspaces in language models (LLMs) can steer scientific code generation toward a specific programming language. Five causal LLMs were first evaluated on scientific coding prompts to quantify their baseline bias among four programming languages. A static neuron-attribution method, perturbing the highest activated MLP weight for a C++ or CPP token, proved brittle and exhibited limited generalization across prompt styles and model scales. To address these limitations, a gradient-refined adaptive activation steering framework (G-ACT) was developed: per-prompt activation differences are clustered into a small set of steering directions, and lightweight per-layer probes are trained and refined online to select the appropriate steering vector. In LLaMA-3.2 3B, this approach reliably biases generation towards the CPP language by increasing the average probe classification accuracy by 15% and the early layers (0-6) improving the probe classification accuracy by 61.5% compared to the standard ACT framework. For LLaMA-3.3 70B, where attention-head signals become more diffuse, targeted injections at key layers still improve language selection. Although per-layer probing introduces a modest inference overhead, it remains practical by steering only a subset of layers and enables reproducible model behavior. These results demonstrate a scalable, interpretable and efficient mechanism for concept-level control for practical agentic systems.
Large Language Model (LLM)-based agents are increasingly employed to automate complex software engineering tasks such as program repair and issue resolution. These agents operate by autonomously generating natural language thoughts, invoking external tools, and iteratively refining their solutions. Despite their widespread adoption, the internal decision-making processes of these agents remain largely unexplored, limiting our understanding of their operational dynamics and failure modes. In this paper, we present a large-scale empirical study of the thought-action-result trajectories of three state-of-the-art LLM-based agents: \textsc{RepairAgent}, \textsc{AutoCodeRover}, and \textsc{OpenHands}. We unify their interaction logs into a common format, capturing 120 trajectories and 2822 LLM interactions focused on program repair and issue resolution. Our study combines quantitative analyses of structural properties, action patterns, and token usage with qualitative assessments of reasoning coherence and feedback integration. We identify key trajectory characteristics such as iteration counts and token consumption, recurring action sequences, and the semantic coherence linking thoughts, actions, and their results. Our findings reveal behavioral motifs and anti-patterns that distinguish successful from failed executions, providing actionable insights for improving agent design, including prompting strategies, failure diagnosis, and anti-pattern detection. We release our dataset and annotation framework to support further research on transparent and robust autonomous software engineering agents.
TRIZ, the Theory of Inventive Problem Solving, is a structured, knowledge-based framework for innovation and abstracting problems to find inventive solutions. However, its application is often limited by the complexity and deep interdisciplinary knowledge required. Advancements in Large Language Models (LLMs) have revealed new possibilities for automating parts of this process. While previous studies have explored single LLMs in TRIZ applications, this paper introduces a multi-agent approach. We propose an LLM-based multi-agent system, called TRIZ agents, each with specialized capabilities and tool access, collaboratively solving inventive problems based on the TRIZ methodology. This multi-agent system leverages agents with various domain expertise to efficiently navigate TRIZ steps. The aim is to model and simulate an inventive process with language agents. We assess the effectiveness of this team of agents in addressing complex innovation challenges based on a selected case study in engineering. We demonstrate the potential of agent collaboration to produce diverse, inventive solutions. This research contributes to the future of AI-driven innovation, showcasing the advantages of decentralized problem-solving in complex ideation tasks.
A potential concern when simulating populations of large language models (LLMs) is data contamination, i.e. the possibility that training data may shape outcomes in unintended ways. While this concern is important and may hinder certain experiments with multi-agent models, it does not preclude the study of genuinely emergent dynamics in LLM populations. The recent critique by Barrie and T\"ornberg [1] of the results of Flint Ashery et al. [2] offers an opportunity to clarify that self-organisation and model-dependent emergent dynamics can be studied in LLM populations, highlighting how such dynamics have been empirically observed in the specific case of social conventions.
In the design process of the analog circuit pre-layout phase, device sizing is an important step in determining whether an analog circuit can meet the required performance metrics. Many existing techniques extract the circuit sizing task as a mathematical optimization problem to solve and continuously improve the optimization efficiency from a mathematical perspective. But they ignore the automatic introduction of prior knowledge, fail to achieve effective pruning of the search space, which thereby leads to a considerable compression margin remaining in the search space. To alleviate this problem, we propose a large language model (LLM)-based multi-agent framework for analog circuits' sizing relationships extraction from academic papers. The search space in the sizing process can be effectively pruned based on the sizing relationship extracted by this framework. Eventually, we conducted tests on 3 types of circuits, and the optimization efficiency was improved by $2.32 \sim 26.6 \times$. This work demonstrates that the LLM can effectively prune the search space for analog circuit sizing, providing a new solution for the combination of LLMs and conventional analog circuit design automation methods.
Scientific progress increasingly relies on effective collaboration among researchers, a dynamic that large language models (LLMs) have only begun to emulate. While recent LLM-based scientist agents show promise in autonomous scientific discovery, they often lack the interactive reasoning and evaluation mechanisms essential to real-world research. We propose IDVSCI (Internal Discussion and Vote SCIentists), a multi-agent framework built on LLMs that incorporates two key innovations: a Dynamic Knowledge Exchange mechanism enabling iterative feedback among agents, and a Dual-Diversity Review paradigm that simulates heterogeneous expert evaluation. These components jointly promote deeper reasoning and the generation of more creative and impactful scientific ideas. To evaluate the effectiveness and generalizability of our approach, we conduct experiments on two datasets: a widely used benchmark in computer science and a new dataset we introduce in the health sciences domain. Results show that IDVSCI consistently achieves the best performance across both datasets, outperforming existing systems such as AI Scientist and VIRSCI. These findings highlight the value of modeling interaction and peer review dynamics in LLM-based autonomous research.
Large Language Models (LLMs) excel at code generation, but ensuring their outputs to be functionally correct, especially in complex programming tasks, is a persistent challenge. While traditional Test-Driven Development (TDD) offers a path for code refinement, its efficacy with LLMs is often undermined by the scarcity of high-quality test cases or the pitfalls of automated test generation, including biased tests or inaccurate output predictions that can misdirect the correction process. This paper introduces Property-Generated Solver, a novel framework that leverages Property-Based Testing (PBT) to validate high-level program properties or invariants, instead of relying on specific input-output examples. These properties are often simpler to define and verify than directly predicting exhaustive test oracles, breaking the "cycle of self-deception" where tests might share flaws with the code they are meant to validate. Property-Generated Solver employs two collaborative LLM-based agents: a Generator dedicated to code generation and iterative refinement, and a Tester that manages the PBT life-cycle and formulate semantically rich feedback from property violations. The resulting comprehensive and actionable feedback then guides the Generator in its refinement efforts. By establishing PBT as the core validation engine within this iterative, closed-loop paradigm, Property-Generated Solver provides a robust mechanism for steering LLMs towards more correct and generalizable code. Extensive experimental results on multiple code generation benchmarks demonstrate that Property-Generated Solver achieves substantial pass@1 improvements, ranging from 23.1% to 37.3% relative gains over established TDD methods.
Academic writing skills are crucial for students' success, but can feel overwhelming without proper guidance and practice, particularly when writing in a second language. Traditionally, students ask instructors or search dictionaries, which are not universally accessible. Early writing assistants emerged as rule-based systems that focused on detecting misspellings, subject-verb disagreements, and basic punctuation errors; however, they are inaccurate and lack contextual understanding. Machine learning-based assistants demonstrate a strong ability for language understanding but are expensive to train. Large language models (LLMs) have shown remarkable capabilities in generating responses in natural languages based on given prompts. Still, they have a fundamental limitation in education: they generate essays without teaching, which can have detrimental effects on learning when misused. To address this limitation, we develop CoachGPT, which leverages large language models (LLMs) to assist individuals with limited educational resources and those who prefer self-paced learning in academic writing. CoachGPT is an AI agent-based web application that (1) takes instructions from experienced educators, (2) converts instructions into sub-tasks, and (3) provides real-time feedback and suggestions using large language models. This unique scaffolding structure makes CoachGPT unique among existing writing assistants. Compared to existing writing assistants, CoachGPT provides a more immersive writing experience with personalized feedback and guidance. Our user studies prove the usefulness of CoachGPT and the potential of large language models for academic writing.
The rapid progress of Large Language Models (LLMs) has given rise to a new category of autonomous AI systems, referred to as Deep Research (DR) agents. These agents are designed to tackle complex, multi-turn informational research tasks by leveraging a combination of dynamic reasoning, adaptive long-horizon planning, multi-hop information retrieval, iterative tool use, and the generation of structured analytical reports. In this paper, we conduct a detailed analysis of the foundational technologies and architectural components that constitute Deep Research agents. We begin by reviewing information acquisition strategies, contrasting API-based retrieval methods with browser-based exploration. We then examine modular tool-use frameworks, including code execution, multimodal input processing, and the integration of Model Context Protocols (MCPs) to support extensibility and ecosystem development. To systematize existing approaches, we propose a taxonomy that differentiates between static and dynamic workflows, and we classify agent architectures based on planning strategies and agent composition, including single-agent and multi-agent configurations. We also provide a critical evaluation of current benchmarks, highlighting key limitations such as restricted access to external knowledge, sequential execution inefficiencies, and misalignment between evaluation metrics and the practical objectives of DR agents. Finally, we outline open challenges and promising directions for future research. A curated and continuously updated repository of DR agent research is available at: {https://github.com/ai-agents-2030/awesome-deep-research-agent}.
AI agents have experienced a paradigm shift, from early dominance by reinforcement learning (RL) to the rise of agents powered by large language models (LLMs), and now further advancing towards a synergistic fusion of RL and LLM capabilities. This progression has endowed AI agents with increasingly strong abilities. Despite these advances, to accomplish complex real-world tasks, agents are required to plan and execute effectively, maintain reliable memory, and coordinate smoothly with other agents. Achieving these capabilities involves contending with ever-present intricate information, operations, and interactions. In light of this challenge, data structurization can play a promising role by transforming intricate and disorganized data into well-structured forms that agents can more effectively understand and process. In this context, graphs, with their natural advantage in organizing, managing, and harnessing intricate data relationships, present a powerful data paradigm for structurization to support the capabilities demanded by advanced AI agents. To this end, this survey presents a first systematic review of how graphs can empower AI agents. Specifically, we explore the integration of graph techniques with core agent functionalities, highlight notable applications, and identify prospective avenues for future research. By comprehensively surveying this burgeoning intersection, we hope to inspire the development of next-generation AI agents equipped to tackle increasingly sophisticated challenges with graphs. Related resources are collected and continuously updated for the community in the Github link.
The rapid spread of misinformation in the digital era poses significant challenges to public discourse, necessitating robust and scalable fact-checking solutions. Traditional human-led fact-checking methods, while credible, struggle with the volume and velocity of online content, prompting the integration of automated systems powered by Large Language Models (LLMs). However, existing automated approaches often face limitations, such as handling complex claims, ensuring source credibility, and maintaining transparency. This paper proposes a novel multi-agent system for automated fact-checking that enhances accuracy, efficiency, and explainability. The system comprises four specialized agents: an Input Ingestion Agent for claim decomposition, a Query Generation Agent for formulating targeted subqueries, an Evidence Retrieval Agent for sourcing credible evidence, and a Verdict Prediction Agent for synthesizing veracity judgments with human-interpretable explanations. Evaluated on benchmark datasets (FEVEROUS, HOVER, SciFact), the proposed system achieves a 12.3% improvement in Macro F1-score over baseline methods. The system effectively decomposes complex claims, retrieves reliable evidence from trusted sources, and generates transparent explanations for verification decisions. Our approach contributes to the growing field of automated fact-checking by providing a more accurate, efficient, and transparent verification methodology that aligns with human fact-checking practices while maintaining scalability for real-world applications. Our source code is available at https://github.com/HySonLab/FactAgent
Large Language Models (LLMs) are being increasingly used in software engineering tasks, with an increased focus on bug report resolution over the past year. However, most proposed systems fail to properly handle uncertain or incorrect inputs and outputs. Existing LLM-based tools and coding agents respond to every issue and generate a patch for every case, even when the input is vague or their own output is incorrect. There are no mechanisms in place to abstain when confidence is low. This leads to unreliable behaviour, such as hallucinated code changes or responses based on vague issue reports. We introduce BouncerBench, a benchmark that evaluates whether LLM-based software agents can refuse to act when inputs are ill-defined or refuse to respond when their own outputs are likely to be incorrect. Unlike prior benchmarks that implicitly incentivize models to generate responses even when uncertain, BouncerBench aims to improve precision by targeting two overlooked failure points: (1) vague or underspecified issue descriptions in tickets and (2) logically or functionally incorrect code patches created by the system. It measures whether proposed systems can distinguish actionable issues from vague tickets and valid patches from untrustworthy ones. We also implement a basic input and output bouncer, evaluating how well current LLMs can abstain when needed. Our results show that most models fail to abstain from underspecified inputs or incorrect outputs. Hence, we conclude that there is significant room for improvement before LLMs can be trusted to make correct decisions and recommendations in real-world software engineering workflows. BouncerBench provides a first step toward evaluating and building more cautious, trustworthy code agents. The replication package, dataset, and leaderboard can be found at bouncerbench.com
Social reasoning - inferring unobservable beliefs and intentions from partial observations of other agents - remains a challenging task for large language models (LLMs). We evaluate the limits of current reasoning language models in the social deduction game Avalon and find that while the largest models demonstrate strong performance, they require extensive test-time inference and degrade sharply when distilled to smaller, real-time-capable variants. To address this, we introduce a hybrid reasoning framework that externalizes belief inference to a structured probabilistic model, while using an LLM for language understanding and interaction. Our approach achieves competitive performance with much larger models in Agent-Agent play and, notably, is the first language agent to defeat human players in a controlled study - achieving a 67% win rate and receiving higher qualitative ratings than both reasoning baselines and human teammates. We release code, models, and a dataset to support future work on social reasoning in LLM agents, which can be found at https://camp-lab-purdue.github.io/bayesian-social-deduction/
Recent progress in large language model (LLM)-based multi-agent collaboration highlights the power of structured communication in enabling collective intelligence. However, existing methods largely rely on static or graph-based inter-agent topologies, lacking the potential adaptability and flexibility in communication. In this work, we propose a new framework that rethinks multi-agent coordination through a sequential structure rather than a graph structure, offering a significantly larger topology space for multi-agent communication. Our method focuses on two key directions: (1) Next-Agent Prediction, which selects the most suitable agent role at each step, and (2) Next-Context Selection (NCS), which enables each agent to selectively access relevant information from any previous step. Together, these components construct task-adaptive communication pipelines that support both role flexibility and global information flow. Extensive evaluations across multiple benchmarks demonstrate that our approach achieves superior performance while substantially reducing communication overhead.
LLM Agents produce patches automatically to resolve an issue. However, they can generate inaccurate patches. Little is known about the root causes behind those failed patches or how those could be fixed. This paper reports an empirical study of the failed patches generated by seven top LLM code agents. We collected 114 issues from the SWE-bench Lite dataset that remained unresolved across the agents. The seven agents produced a total of 769 failed patches for those issues, which we checked with a combination of GPT-4o and manual analysis. We present a taxonomy of the failure reasons across the patches. The taxonomy contains six categories, with several sub-categories under each category. For example, a frequently observed category is the inability of an LLM to correctly infer/produce the appropriate variable type in the produced patch. As a first step towards addressing such type-related errors, we designed PAGENT (Patch Agent). PAGENT utilizes program analysis techniques like CFG creation and exploration to infer the type of information of a patch. PAGENT does this by applying repository-level static code analysis techniques. Then, PAGENT refines the inferred type by further utilizing an LLM-based inference technique. We tested PAGENT on all 127 type-related failed patches from the top three agents in our study. PAGENT could fix 29 of the 127 failed patches.
Current recommendation systems often require some form of textual data summarization, such as generating concise and coherent titles for product carousels or other grouped item displays. While large language models have shown promise in NLP domains for textual summarization, these approaches do not directly apply to recommendation systems, where explanations must be highly relevant to the core features of item sets, adhere to strict word limit constraints. In this paper, we propose CARTS (Collaborative Agents for Recommendation Textual Summarization), a multi-agent LLM framework designed for structured summarization in recommendation systems. CARTS decomposes the task into three stages-Generation Augmented Generation (GAG), refinement circle, and arbitration, where successive agent roles are responsible for extracting salient item features, iteratively refining candidate titles based on relevance and length feedback, and selecting the final title through a collaborative arbitration process. Experiments on large-scale e-commerce data and live A/B testing show that CARTS significantly outperforms single-pass and chain-of-thought LLM baselines, delivering higher title relevance and improved user engagement metrics.
The advent of Large Language Models (LLMs) enables the rise of App agents that interpret user intent and operate smartphone Apps through actions such as clicking and scrolling. While prompt-based solutions with closed LLM APIs show promising ability, they incur heavy compute costs and external API dependency. Fine-tuning smaller open-source LLMs solves these limitations. However, current fine-tuning methods use a syntax learning paradigm that forces agents to reproduce exactly the ground truth action strings, leading to out-of-distribution (OOD) vulnerability. To fill this gap, we propose Action Semantics Learning (ASL), a novel learning framework, where the learning objective is capturing the semantics of the ground truth actions. Specifically, inspired by the programming language theory, we define the action semantics for App agents as the state transition induced by the action in the user interface. With this insight, ASL employs a novel SEmantic Estimator (SEE) to compute a semantic reward to train the App agents in generating actions aligned with the semantics of ground truth actions, even when the syntactic forms differ. To support the effectiveness of ASL, we theoretically demonstrate the superior robustness of ASL for the OOD problem compared with the existing syntax learning paradigm. Extensive experiments on offline and online smartphone App operation benchmarks show that ASL significantly improves the accuracy and generalisation of App agents over existing methods.
Artificial Intelligence (AI) Infrastructures, represented by Deep Learning (DL) frameworks, have served as fundamental DL systems over the last decade. However, the bugs in DL frameworks could lead to catastrophic consequences in some critical scenarios (e.g., healthcare and autonomous driving). A simple yet effective way to find bugs in DL frameworks is fuzz testing (Fuzzing). Unfortunately, existing fuzzing techniques have not comprehensively considered multiple types of feedback. Additionally, they analyze feedback in a coarse-grained manner, such as mutating the test cases only according to whether the coverage increases. Recently, researchers introduced Large Language Models (LLMs) into fuzzing. However, current LLM-based fuzzing techniques only focus on using LLMs to generate test cases while overlooking their potential to analyze feedback information, failing to create more valid and diverse test cases. To fill this gap, we propose FUEL to break the seal of Feedback-driven fuzzing for DL frameworks. The backbone of FUEL comprises two LLM-based agents, namely analysis LLM and generation LLM. Analysis LLM agent infers analysis summaries from feedback information, while the generation LLM agent creates tests guided by these analysis summaries. So far, FUEL has detected 104 bugs for PyTorch and TensorFlow, with 93 confirmed as new bugs, 47 already fixed, and 5 assigned with CVE IDs. Our work indicates that considering multiple types of feedback is beneficial to fuzzing performance, and leveraging LLMs to analyze feedback information is a promising direction. Our artifact is available at https://github.com/NJU-iSE/FUEL
The growing dependence on mobile phones and their apps has made multi-user interactive features, like chat calls, live streaming, and video conferencing, indispensable for bridging the gaps in social connectivity caused by physical and situational barriers. However, automating these interactive features for testing is fraught with challenges, owing to their inherent need for timely, dynamic, and collaborative user interactions, which current automated testing methods inadequately address. Inspired by the concept of agents designed to autonomously and collaboratively tackle problems, we propose MAdroid, a novel multi-agent approach powered by the Large Language Models (LLMs) to automate the multi-user interactive task for app feature testing. Specifically, MAdroid employs two functional types of multi-agents: user agents (Operator) and supervisor agents (Coordinator and Observer). Each agent takes a specific role: the Coordinator directs the interactive task; the Operator mimics user interactions on the device; and the Observer monitors and reviews the task automation process. Our evaluation, which included 41 multi-user interactive tasks, demonstrates the effectiveness of our approach, achieving 82.9% of the tasks with 96.8% action similarity, outperforming the ablation studies and state-of-the-art baselines. Additionally, a preliminary investigation underscores MAdroid's practicality by helping identify 11 multi-user interactive bugs during regression app testing, confirming its potential value in real-world software development contexts.
Supply chain operations generate vast amounts of operational data; however, critical knowledge such as system usage practices, troubleshooting workflows, and resolution techniques often remains buried within unstructured communications like support tickets, emails, and chat logs. While RAG systems aim to leverage such communications as a knowledge base, their effectiveness is limited by raw data challenges: support tickets are typically noisy, inconsistent, and incomplete, making direct retrieval suboptimal. Unlike existing RAG approaches that focus on runtime optimization, we introduce a novel offline-first methodology that transforms these communications into a structured knowledge base. Our key innovation is a LLMs-based multi-agent system orchestrating three specialized agents: Category Discovery for taxonomy creation, Categorization for ticket grouping, and Knowledge Synthesis for article generation. Applying our methodology to real-world support tickets with resolution notes and comments, our system creates a compact knowledge base - reducing total volume to just 3.4% of original ticket data while improving quality. Experiments demonstrate that our prebuilt knowledge base in RAG systems significantly outperforms traditional RAG implementations (48.74% vs. 38.60% helpful answers) and achieves a 77.4% reduction in unhelpful responses. By automating institutional knowledge capture that typically remains siloed in experts' heads, our solution translates to substantial operational efficiency: reducing support workload, accelerating resolution times, and creating self-improving systems that automatically resolve approximately 50% of future supply chain tickets. Our approach addresses a key gap in knowledge management by transforming transient communications into structured, reusable knowledge through intelligent offline processing rather than latency-inducing runtime architectures.
Efforts to improve Large Language Model (LLM) agent performance on complex tasks have largely focused on fine-tuning and iterative self-correction. However, these approaches often lack generalizable mechanisms for longterm learning and remain inefficient in dynamic environments. We introduce OmniReflect, a hierarchical, reflection-driven framework that constructs a constitution, a compact set of guiding principles distilled from task experiences, to enhance the effectiveness and efficiency of an LLM agent. OmniReflect operates in two modes: Self-sustaining, where a single agent periodically curates its own reflections during task execution, and Co-operative, where a Meta-advisor derives a constitution from a small calibration set to guide another agent. To construct these constitutional principles, we employ Neural, Symbolic, and NeuroSymbolic techniques, offering a balance between contextual adaptability and computational efficiency. Empirical results averaged across models show major improvements in task success, with absolute gains of +10.3% on ALFWorld, +23.8% on BabyAI, and +8.3% on PDDL in the Self-sustaining mode. Similar gains are seen in the Co-operative mode, where a lightweight Qwen3-4B ReAct agent outperforms all Reflexion baselines on BabyAI. These findings highlight the robustness and effectiveness of OmniReflect across environments and backbones.
As Large Language Models (LLMs) are integrated into safety-critical applications involving sequential decision-making in the real world, it is essential to know when to trust LLM decisions. Existing LLM Uncertainty Quantification (UQ) methods are primarily designed for single-turn question-answering formats, resulting in multi-step decision-making scenarios, e.g., LLM agentic system, being underexplored. In this paper, we introduce a principled, information-theoretic framework that decomposes LLM sequential decision uncertainty into two parts: (i) internal uncertainty intrinsic to the current decision, which is focused on existing UQ methods, and (ii) extrinsic uncertainty, a Mutual-Information (MI) quantity describing how much uncertainty should be inherited from preceding decisions. We then propose UProp, an efficient and effective extrinsic uncertainty estimator that converts the direct estimation of MI to the estimation of Pointwise Mutual Information (PMI) over multiple Trajectory-Dependent Decision Processes (TDPs). UProp is evaluated over extensive multi-step decision-making benchmarks, e.g., AgentBench and HotpotQA, with state-of-the-art LLMs, e.g., GPT-4.1 and DeepSeek-V3. Experimental results demonstrate that UProp significantly outperforms existing single-turn UQ baselines equipped with thoughtful aggregation strategies. Moreover, we provide a comprehensive analysis of UProp, including sampling efficiency, potential applications, and intermediate uncertainty propagation, to demonstrate its effectiveness. Codes will be available at https://github.com/jinhaoduan/UProp.
The rapid progress in Automated Program Repair (APR) has been driven by advances in AI, particularly large language models (LLMs) and agent-based systems. SWE-Bench is a recent benchmark designed to evaluate LLM-based repair systems using real issues and pull requests mined from 12 popular open-source Python repositories. Its public leaderboards, SWE-Bench Lite and SWE-Bench Verified, have become central platforms for tracking progress and comparing solutions. However, because the submission process does not require detailed documentation, the architectural design and origin of many solutions remain unclear. In this paper, we present the first comprehensive study of all submissions to the SWE-Bench Lite (68 entries) and Verified (79 entries) leaderboards, analyzing 67 unique approaches across dimensions such as submitter type, product availability, LLM usage, and system architecture. Our findings reveal the dominance of proprietary LLMs (especially Claude 3.5/3.7), the presence of both agentic and non-agentic designs, and a contributor base spanning from individual developers to large tech companies.
In this paper, we introduce the AI Search Paradigm, a comprehensive blueprint for next-generation search systems capable of emulating human information processing and decision-making. The paradigm employs a modular architecture of four LLM-powered agents (Master, Planner, Executor and Writer) that dynamically adapt to the full spectrum of information needs, from simple factual queries to complex multi-stage reasoning tasks. These agents collaborate dynamically through coordinated workflows to evaluate query complexity, decompose problems into executable plans, and orchestrate tool usage, task execution, and content synthesis. We systematically present key methodologies for realizing this paradigm, including task planning and tool integration, execution strategies, aligned and robust retrieval-augmented generation, and efficient LLM inference, spanning both algorithmic techniques and infrastructure-level optimizations. By providing an in-depth guide to these foundational components, this work aims to inform the development of trustworthy, adaptive, and scalable AI search systems.
Recent work on large language models has demonstrated the use of model-free reinforcement learning (RL) to train reasoning-like capabilities. The emergence of "thinking" through model-free RL is interesting as thinking actions neither produce reward nor change the external world state to one where the agent is more likely to get reward. This paper seeks to build a domain-independent understanding of when model-free RL will lead to "thinking" as a strategy for reward maximization. To build this understanding, we first introduce a theoretical model which we call a \textit{thought Markov decision process} (MDP). Thought MDPs minimally extend the classical MDP model to include an abstract notion of thought state and thought action. Using the thought MDP model, we prove the importance of policy initialization in determining whether or not thinking emerges and show formally that thought actions are equivalent to the agent choosing to perform a step of policy improvement before continuing to act. We then show that open-source LLMs satisfy the conditions that our theory predicts are necessary for model-free RL to produce thinking-like behavior. Finally, we hypothesize sufficient conditions that would enable thinking to be learned outside of language generation and introduce a toy domain where a combination of multi-task pre-training and designated thought actions enable more data-efficient RL compared to non-thinking agents.
Large Language Models (LLMs) are increasingly proposed as near-autonomous artificial intelligence (AI) agents capable of making everyday decisions on behalf of humans. Although LLMs perform well on many technical tasks, their behaviour in personal decision-making remains less understood. Previous studies have assessed their rationality and moral alignment with human decisions. However, the behaviour of AI assistants in scenarios where financial rewards are at odds with user comfort has not yet been thoroughly explored. In this paper, we tackle this problem by quantifying the prices assigned by multiple LLMs to a series of user discomforts: additional walking, waiting, hunger and pain. We uncover several key concerns that strongly question the prospect of using current LLMs as decision-making assistants: (1) a large variance in responses between LLMs, (2) within a single LLM, responses show fragility to minor variations in prompt phrasing (e.g., reformulating the question in the first person can considerably alter the decision), (3) LLMs can accept unreasonably low rewards for major inconveniences (e.g., 1 Euro to wait 10 hours), and (4) LLMs can reject monetary gains where no discomfort is imposed (e.g., 1,000 Euro to wait 0 minutes). These findings emphasize the need for scrutiny of how LLMs value human inconvenience, particularly as we move toward applications where such cash-versus-comfort trade-offs are made on users' behalf.
Traditional technical analysis methods face limitations in accurately predicting trends in today's complex financial markets. This paper introduces ElliottAgents, an multi-agent system that integrates the Elliott Wave Principle with AI for stock market forecasting. The inherent complexity of financial markets, characterized by non-linear dynamics, noise, and susceptibility to unpredictable external factors, poses significant challenges for accurate prediction. To address these challenges, the system employs LLMs to enhance natural language understanding and decision-making capabilities within a multi-agent framework. By leveraging technologies such as Retrieval-Augmented Generation (RAG) and Deep Reinforcement Learning (DRL), ElliottAgents performs continuous, multi-faceted analysis of market data to identify wave patterns and predict future price movements. The research explores the system's ability to process historical stock data, recognize Elliott wave patterns, and generate actionable insights for traders. Experimental results, conducted on historical data from major U.S. companies, validate the system's effectiveness in pattern recognition and trend forecasting across various time frames. This paper contributes to the field of AI-driven financial analysis by demonstrating how traditional technical analysis methods can be effectively combined with modern AI approaches to create more reliable and interpretable market prediction systems.
Intelligent Transportation Systems (ITS) are increasingly vulnerable to sophisticated cyberattacks due to their complex, interconnected nature. Ensuring the cybersecurity of these systems is paramount to maintaining road safety and minimizing traffic disruptions. This study presents a novel multi-agent framework leveraging Large Language Models (LLMs) to enhance traffic simulation and cybersecurity testing. The framework automates the creation of traffic scenarios, the design of cyberattack strategies, and the development of defense mechanisms. A case study demonstrates the framework's ability to simulate a cyberattack targeting connected vehicle broadcasts, evaluate its impact, and implement a defense mechanism that significantly mitigates traffic delays. Results show a 10.2 percent increase in travel time during an attack, which is reduced by 3.3 percent with the defense strategy. This research highlights the potential of LLM-driven multi-agent systems in advancing transportation cybersecurity and offers a scalable approach for future research in traffic simulation and cyber defense.
Large Language Models (LLMs) have shown impressive capabilities in downstream software engineering tasks such as Automated Program Repair (APR). In particular, there has been a lot of research on repository-level issue-resolution benchmarks such as SWE-Bench. Although there has been significant progress on this topic, we notice that in the process of solving such issues, existing agentic systems tend to hyper-localize on immediately suspicious lines of code and fix them in isolation, without a deeper understanding of the issue semantics, code semantics, or execution semantics. Consequently, many existing systems generate patches that overfit to the user issue, even when a more general fix is preferable. To address this limitation, we introduce SemAgent, a novel workflow-based procedure that leverages issue, code, and execution semantics to generate patches that are complete - identifying and fixing all lines relevant to the issue. We achieve this through a novel pipeline that (a) leverages execution semantics to retrieve relevant context, (b) comprehends issue-semantics via generalized abstraction, (c) isolates code-semantics within the context of this abstraction, and (d) leverages this understanding in a two-stage architecture: a repair stage that proposes fine-grained fixes, followed by a reviewer stage that filters relevant fixes based on the inferred issue-semantics. Our evaluations show that our methodology achieves a solve rate of 44.66% on the SWEBench-Lite benchmark beating all other workflow-based approaches, and an absolute improvement of 7.66% compared to our baseline, which lacks such deep semantic understanding. We note that our approach performs particularly well on issues requiring multi-line reasoning (and editing) and edge-case handling, suggesting that incorporating issue and code semantics into APR pipelines can lead to robust and semantically consistent repairs.
Traditional quality assurance (QA) methods face significant challenges in addressing the complexity, scale, and rapid iteration cycles of modern software systems and are strained by limited resources available, leading to substantial costs associated with poor quality. The object of this research is the Quality Assurance processes for modern distributed software applications. The subject of the research is the assessment of the benefits, challenges, and prospects of integrating modern AI-oriented tools into quality assurance processes. We performed comprehensive analysis of implications on both verification and validation processes covering exploratory test analyses, equivalence partitioning and boundary analyses, metamorphic testing, finding inconsistencies in acceptance criteria (AC), static analyses, test case generation, unit test generation, test suit optimization and assessment, end to end scenario execution. End to end regression of sample enterprise application utilizing AI-agents over generated test scenarios was implemented as a proof of concept highlighting practical use of the study. The results, with only 8.3% flaky executions of generated test cases, indicate significant potential for the proposed approaches. However, the study also identified substantial challenges for practical adoption concerning generation of semantically identical coverage, "black box" nature and lack of explainability from state-of-the-art Large Language Models (LLMs), the tendency to correct mutated test cases to match expected results, underscoring the necessity for thorough verification of both generated artifacts and test execution results. The research demonstrates AI's transformative potential for QA but highlights the importance of a strategic approach to implementing these technologies, considering the identified limitations and the need for developing appropriate verification methodologies.