TextArena is an open-source collection of competitive text-based games for training and evaluation of agentic behavior in Large Language Models (LLMs). It spans 57+ unique environments (including single-player, two-player, and multi-player setups) and allows for easy evaluation of model capabilities via an online-play system (against humans and other submitted models) with real-time TrueSkill scores. Traditional benchmarks rarely assess dynamic social skills such as negotiation, theory of mind, and deception, creating a gap that TextArena addresses. Designed with research, community and extensibility in mind, TextArena emphasizes ease of adding new games, adapting the framework, testing models, playing against the models, and training models. Detailed documentation of environments, games, leaderboard, and examples are available on https://github.com/LeonGuertler/TextArena and https://www.textarena.ai/.
Cancer patients are increasingly turning to large language models (LLMs) as a new form of internet search for medical information, making it critical to assess how well these models handle complex, personalized questions. However, current medical benchmarks focus on medical exams or consumer-searched questions and do not evaluate LLMs on real patient questions with detailed clinical contexts. In this paper, we first evaluate LLMs on cancer-related questions drawn from real patients, reviewed by three hematology oncology physicians. While responses are generally accurate, with GPT-4-Turbo scoring 4.13 out of 5, the models frequently fail to recognize or address false presuppositions in the questions-posing risks to safe medical decision-making. To study this limitation systematically, we introduce Cancer-Myth, an expert-verified adversarial dataset of 585 cancer-related questions with false presuppositions. On this benchmark, no frontier LLM -- including GPT-4o, Gemini-1.Pro, and Claude-3.5-Sonnet -- corrects these false presuppositions more than 30% of the time. Even advanced medical agentic methods do not prevent LLMs from ignoring false presuppositions. These findings expose a critical gap in the clinical reliability of LLMs and underscore the need for more robust safeguards in medical AI systems.
LLM-integrated applications and agents are vulnerable to prompt injection attacks, where an attacker injects prompts into their inputs to induce attacker-desired outputs. A detection method aims to determine whether a given input is contaminated by an injected prompt. However, existing detection methods have limited effectiveness against state-of-the-art attacks, let alone adaptive ones. In this work, we propose DataSentinel, a game-theoretic method to detect prompt injection attacks. Specifically, DataSentinel fine-tunes an LLM to detect inputs contaminated with injected prompts that are strategically adapted to evade detection. We formulate this as a minimax optimization problem, with the objective of fine-tuning the LLM to detect strong adaptive attacks. Furthermore, we propose a gradient-based method to solve the minimax optimization problem by alternating between the inner max and outer min problems. Our evaluation results on multiple benchmark datasets and LLMs show that DataSentinel effectively detects both existing and adaptive prompt injection attacks.
Large Language Model (LLM)-based agents have demonstrated strong capabilities across a wide range of tasks, and their application in the medical domain holds particular promise due to the demand for high generalizability and reliance on interdisciplinary knowledge. However, existing medical agent systems often rely on static, manually crafted workflows that lack the flexibility to accommodate diverse diagnostic requirements and adapt to emerging clinical scenarios. Motivated by the success of automated machine learning (AutoML), this paper introduces a novel framework for the automated design of medical agent architectures. Specifically, we define a hierarchical and expressive agent search space that enables dynamic workflow adaptation through structured modifications at the node, structural, and framework levels. Our framework conceptualizes medical agents as graph-based architectures composed of diverse, functional node types and supports iterative self-improvement guided by diagnostic feedback. Experimental results on skin disease diagnosis tasks demonstrate that the proposed method effectively evolves workflow structures and significantly enhances diagnostic accuracy over time. This work represents the first fully automated framework for medical agent architecture design and offers a scalable, adaptable foundation for deploying intelligent agents in real-world clinical environments.
A Large Language Model (LLM) powered GUI agent is a specialized autonomous system that performs tasks on the user's behalf according to high-level instructions. It does so by perceiving and interpreting the graphical user interfaces (GUIs) of relevant apps, often visually, inferring necessary sequences of actions, and then interacting with GUIs by executing the actions such as clicking, typing, and tapping. To complete real-world tasks, such as filling forms or booking services, GUI agents often need to process and act on sensitive user data. However, this autonomy introduces new privacy and security risks. Adversaries can inject malicious content into the GUIs that alters agent behaviors or induces unintended disclosures of private information. These attacks often exploit the discrepancy between visual saliency for agents and human users, or the agent's limited ability to detect violations of contextual integrity in task automation. In this paper, we characterized six types of such attacks, and conducted an experimental study to test these attacks with six state-of-the-art GUI agents, 234 adversarial webpages, and 39 human participants. Our findings suggest that GUI agents are highly vulnerable, particularly to contextually embedded threats. Moreover, human users are also susceptible to many of these attacks, indicating that simple human oversight may not reliably prevent failures. This misalignment highlights the need for privacy-aware agent design. We propose practical defense strategies to inform the development of safer and more reliable GUI agents.
We study the automated derivation of safety requirements in a self-driving vehicle use case, leveraging LLMs in combination with agent-based retrieval-augmented generation. Conventional approaches that utilise pre-trained LLMs to assist in safety analyses typically lack domain-specific knowledge. Existing RAG approaches address this issue, yet their performance deteriorates when handling complex queries and it becomes increasingly harder to retrieve the most relevant information. This is particularly relevant for safety-relevant applications. In this paper, we propose the use of agent-based RAG to derive safety requirements and show that the retrieved information is more relevant to the queries. We implement an agent-based approach on a document pool of automotive standards and the Apollo case study, as a representative example of an automated driving perception system. Our solution is tested on a data set of safety requirement questions and answers, extracted from the Apollo data. Evaluating a set of selected RAG metrics, we present and discuss advantages of a agent-based approach compared to default RAG methods.
The fusion of Large Language Models (LLMs) with recommender systems (RecSys) has dramatically advanced personalized recommendations and drawn extensive attention. Despite the impressive progress, the safety of LLM-based RecSys against backdoor attacks remains largely under-explored. In this paper, we raise a new problem: Can a backdoor with a specific trigger be injected into LLM-based Recsys, leading to the manipulation of the recommendation responses when the backdoor trigger is appended to an item's title? To investigate the vulnerabilities of LLM-based RecSys under backdoor attacks, we propose a new attack framework termed Backdoor Injection Poisoning for RecSys (BadRec). BadRec perturbs the items' titles with triggers and employs several fake users to interact with these items, effectively poisoning the training set and injecting backdoors into LLM-based RecSys. Comprehensive experiments reveal that poisoning just 1% of the training data with adversarial examples is sufficient to successfully implant backdoors, enabling manipulation of recommendations. To further mitigate such a security threat, we propose a universal defense strategy called Poison Scanner (P-Scanner). Specifically, we introduce an LLM-based poison scanner to detect the poisoned items by leveraging the powerful language understanding and rich knowledge of LLMs. A trigger augmentation agent is employed to generate diverse synthetic triggers to guide the poison scanner in learning domain-specific knowledge of the poisoned item detection task. Extensive experiments on three real-world datasets validate the effectiveness of the proposed P-Scanner.
Large Language Models (LLMs) have shown outstanding performance across a variety of tasks, partly due to advanced prompting techniques. However, these techniques often require lengthy prompts, which increase computational costs and can hinder performance because of the limited context windows of LLMs. While prompt compression is a straightforward solution, existing methods confront the challenges of retaining essential information, adapting to context changes, and remaining effective across different tasks. To tackle these issues, we propose a task-agnostic method called Dynamic Compressing Prompts (LLM-DCP). Our method reduces the number of prompt tokens while aiming to preserve the performance as much as possible. We model prompt compression as a Markov Decision Process (MDP), enabling the DCP-Agent to sequentially remove redundant tokens by adapting to dynamic contexts and retaining crucial content. We develop a reward function for training the DCP-Agent that balances the compression rate, the quality of the LLM output, and the retention of key information. This allows for prompt token reduction without needing an external black-box LLM. Inspired by the progressive difficulty adjustment in curriculum learning, we introduce a Hierarchical Prompt Compression (HPC) training strategy that gradually increases the compression difficulty, enabling the DCP-Agent to learn an effective compression method that maintains information integrity. Experiments demonstrate that our method outperforms state-of-the-art techniques, especially at higher compression rates. The code for our approach will be available at https://github.com/Fhujinwu/DCP.
This paper presents a realistic simulated stock market where large language models (LLMs) act as heterogeneous competing trading agents. The open-source framework incorporates a persistent order book with market and limit orders, partial fills, dividends, and equilibrium clearing alongside agents with varied strategies, information sets, and endowments. Agents submit standardized decisions using structured outputs and function calls while expressing their reasoning in natural language. Three findings emerge: First, LLMs demonstrate consistent strategy adherence and can function as value investors, momentum traders, or market makers per their instructions. Second, market dynamics exhibit features of real financial markets, including price discovery, bubbles, underreaction, and strategic liquidity provision. Third, the framework enables analysis of LLMs' responses to varying market conditions, similar to partial dependence plots in machine-learning interpretability. The framework allows simulating financial theories without closed-form solutions, creating experimental designs that would be costly with human participants, and establishing how prompts can generate correlated behaviors affecting market stability.
This article explores the growing impact of large language models (LLMs) and Generative AI (GenAI) tools on Internet traffic, focusing on their role as a new and significant source of network load. As these AI tools continue to gain importance in applications ranging from virtual assistants to content generation, the volume of traffic they generate is expected to increase massively. These models use the Internet as the global infrastructure for delivering multimedia messages (text, voice, images, video, etc.) to users, by interconnecting users and devices with AI agents typically deployed in the cloud. We believe this represents a new paradigm that will lead to a considerable increase in network traffic, and network operators must be prepared to address the resulting demands. To support this claim, we provide a proof-of-concept and source code for measuring traffic in remote user-agent interactions, estimating the traffic generated per prompt for some of the most popular open-source LLMs in 2025. The average size of each prompt query and response is 7,593 bytes, with a standard deviation of 369 bytes. These numbers are comparable with email and web browsing traffic. However, we envision AI as the next ``killer application" that will saturate networks with traffic, such as Peer-to-Peer traffic and Video-on-demand dominated in previous decades.
Large language models (LLMs) demonstrate the ability to simulate human decision-making processes, enabling their use as agents in modeling sophisticated social networks, both offline and online. Recent research has explored collective behavioral patterns and structural characteristics of LLM agents within simulated networks. However, empirical comparisons between LLM-driven and human-driven online social networks remain scarce, limiting our understanding of how LLM agents differ from human users. This paper presents a large-scale analysis of Chirper.ai, an X/Twitter-like social network entirely populated by LLM agents, comprising over 65,000 agents and 7.7 million AI-generated posts. For comparison, we collect a parallel dataset from Mastodon, a human-driven decentralized social network, with over 117,000 users and 16 million posts. We examine key differences between LLM agents and humans in posting behaviors, abusive content, and social network structures. Our findings provide critical insights into the evolving landscape of online social network analysis in the AI era, offering a comprehensive profile of LLM agents in social simulations.
Multi-agents-based news-driven time series forecasting is considered as a potential paradigm shift in the era of large language models (LLMs). The challenge of this task lies in measuring the influences of different news events towards the fluctuations of time series. This requires agents to possess stronger abilities of innovative thinking and the identifying misleading logic. However, the existing multi-agent discussion framework has limited enhancement on time series prediction in terms of optimizing these two capabilities. Inspired by the role of competition in fostering innovation, this study embeds a competition mechanism within the multi-agent discussion to enhance agents' capability of generating innovative thoughts. Furthermore, to bolster the model's proficiency in identifying misleading information, we incorporate a fine-tuned small-scale LLM model within the reflective stage, offering auxiliary decision-making support. Experimental results confirm that the competition can boost agents' capacity for innovative thinking, which can significantly improve the performances of time series prediction. Similar to the findings of social science, the intensity of competition within this framework can influence the performances of agents, providing a new perspective for studying LLMs-based multi-agent systems.
Despite the rapid advancement of large language models, they remain highly susceptible to generating hallucinations, which significantly hinders their widespread application. Hallucination research requires dynamic and fine-grained evaluation. However, most existing hallucination benchmarks (especially in Chinese language) rely on human annotations, making automatical and cost-effective hallucination evaluation challenging. To address this, we introduce HaluAgent, an agentic framework that automatically constructs fine-grained QA dataset based on some knowledge documents. Our experiments demonstrate that the manually designed rules and prompt optimization can improve the quality of generated data. Using HaluAgent, we construct C-FAITH, a Chinese QA hallucination benchmark created from 1,399 knowledge documents obtained from web scraping, totaling 60,702 entries. We comprehensively evaluate 16 mainstream LLMs with our proposed C-FAITH, providing detailed experimental results and analysis.
We propose CRAVE (Cluster-based Retrieval Augmented Verification with Explanation); a novel framework that integrates retrieval-augmented Large Language Models (LLMs) with clustering techniques to address fact-checking challenges on social media. CRAVE automatically retrieves multimodal evidence from diverse, often contradictory, sources. Evidence is clustered into coherent narratives, and evaluated via an LLM-based judge to deliver fact-checking verdicts explained by evidence summaries. By synthesizing evidence from both text and image modalities and incorporating agent-based refinement, CRAVE ensures consistency and diversity in evidence representation. Comprehensive experiments demonstrate CRAVE's efficacy in retrieval precision, clustering quality, and judgment accuracy, showcasing its potential as a robust decision-support tool for fact-checkers.
Social simulation is transforming traditional social science research by modeling human behavior through interactions between virtual individuals and their environments. With recent advances in large language models (LLMs), this approach has shown growing potential in capturing individual differences and predicting group behaviors. However, existing methods face alignment challenges related to the environment, target users, interaction mechanisms, and behavioral patterns. To this end, we introduce SocioVerse, an LLM-agent-driven world model for social simulation. Our framework features four powerful alignment components and a user pool of 10 million real individuals. To validate its effectiveness, we conducted large-scale simulation experiments across three distinct domains: politics, news, and economics. Results demonstrate that SocioVerse can reflect large-scale population dynamics while ensuring diversity, credibility, and representativeness through standardized procedures and minimal manual adjustments.
Personalization has become an essential capability in modern AI systems, enabling customized interactions that align with individual user preferences, contexts, and goals. Recent research has increasingly concentrated on Retrieval-Augmented Generation (RAG) frameworks and their evolution into more advanced agent-based architectures within personalized settings to enhance user satisfaction. Building on this foundation, this survey systematically examines personalization across the three core stages of RAG: pre-retrieval, retrieval, and generation. Beyond RAG, we further extend its capabilities into the realm of Personalized LLM-based Agents, which enhance traditional RAG systems with agentic functionalities, including user understanding, personalized planning and execution, and dynamic generation. For both personalization in RAG and agent-based personalization, we provide formal definitions, conduct a comprehensive review of recent literature, and summarize key datasets and evaluation metrics. Additionally, we discuss fundamental challenges, limitations, and promising research directions in this evolving field. Relevant papers and resources are continuously updated at https://github.com/Applied-Machine-Learning-Lab/Awesome-Personalized-RAG-Agent.
Large language models (LLMs) and multimodal large language models (MLLMs) have significantly advanced artificial intelligence. However, visual reasoning, reasoning involving both visual and textual inputs, remains underexplored. Recent advancements, including the reasoning models like OpenAI o1 and Gemini 2.0 Flash Thinking, which incorporate image inputs, have opened this capability. In this ongoing work, we focus specifically on photography-related tasks because a photo is a visual snapshot of the physical world where the underlying physics (i.e., illumination, blur extent, etc.) interplay with the camera parameters. Successfully reasoning from the visual information of a photo to identify these numerical camera settings requires the MLLMs to have a deeper understanding of the underlying physics for precise visual comprehension, representing a challenging and intelligent capability essential for practical applications like photography assistant agents. We aim to evaluate MLLMs on their ability to distinguish visual differences related to numerical camera settings, extending a methodology previously proposed for vision-language models (VLMs). Our preliminary results demonstrate the importance of visual reasoning in photography-related tasks. Moreover, these results show that no single MLLM consistently dominates across all evaluation tasks, demonstrating ongoing challenges and opportunities in developing MLLMs with better visual reasoning.
Large language models (LLMs) have shown promising performance in automated code generation, especially excelling in simple tasks such as generating standalone codes. Different from simple tasks, real-world code generation usually depends on specific programming environment (e.g., code repositories). It contains complex dependencies and domain knowledge, which is needed for LLMs when generating target code snippets. In this paper, we propose CodeRAG, a retrieval-augmented code generation (RAG) framework to comprehensively retrieve supportive codes for real-world code generation. Beginning with the requirement, CodeRAG first constructs a requirement graph for the current repository, and retrieves sub- and similar- requirement nodes of the target requirement on the graph. Meanwhile, it models the repository into a DS-code graph. CodeRAG then maps these relevant requirement nodes into their corresponding code nodes, and treats these code nodes as archors for LLM reasoning on DS-code graph. Finally, CodeRAG introduces a code-oriented agentic reasoning process, seamlessly allowing LLMs to reason and comprehensively retrieve for supportive codes which LLMs' need for generating correct programs. Experiments show that CodeRAG achieves significant improvements (i.e., increasing 40.90 and 37.79 Pass@1 on GPT-4o and Gemini-Pro on DevEval) compared to no RAG scenarios. Further tests on reasoning LLMs (i.e., QwQ-32B) confirm CodeRAG's adaptability and efficacy across various types of LLMs. In addition, CodeRAG outperforms commercial programming products such as Copilit and Cursor. We further investigate the performance of our framework on different dependency types, and observe that CodeRAG is superior in generating examples where target codes invoke predefined cross-file code snippets. These results demonstrate CodeRAG's potential in solving real-world repo-level coding challenges.
Large Language Models (LLMs) are transforming data analytics, but their widespread adoption is hindered by two critical limitations: they are not explainable (opaque reasoning processes) and not verifiable (prone to hallucinations and unchecked errors). While retrieval-augmented generation (RAG) improves accuracy by grounding LLMs in external data, it fails to address the core challenges of trustworthy analytics - especially when processing noisy, inconsistent, or multi-modal data (for example, text, tables, images). We propose DataMosaic, a framework designed to make LLM-powered analytics both explainable and verifiable. By dynamically extracting task-specific structures (for example, tables, graphs, trees) from raw data, DataMosaic provides transparent, step-by-step reasoning traces and enables validation of intermediate results. Built on a multi-agent framework, DataMosaic orchestrates self-adaptive agents that align with downstream task requirements, enhancing consistency, completeness, and privacy. Through this approach, DataMosaic not only tackles the limitations of current LLM-powered analytics systems but also lays the groundwork for a new paradigm of grounded, accurate, and explainable multi-modal data analytics.
Effective pest management is complex due to the need for accurate, context-specific decisions. Recent advancements in large language models (LLMs) open new possibilities for addressing these challenges by providing sophisticated, adaptive knowledge acquisition and reasoning. However, existing LLM-based pest management approaches often rely on a single-agent paradigm, which can limit their capacity to incorporate diverse external information, engage in systematic validation, and address complex, threshold-driven decisions. To overcome these limitations, we introduce PestMA, an LLM-based multi-agent system (MAS) designed to generate reliable and evidence-based pest management advice. Building on an editorial paradigm, PestMA features three specialized agents, an Editor for synthesizing pest management recommendations, a Retriever for gathering relevant external data, and a Validator for ensuring correctness. Evaluations on real-world pest scenarios demonstrate that PestMA achieves an initial accuracy of 86.8% for pest management decisions, which increases to 92.6% after validation. These results underscore the value of collaborative agent-based workflows in refining and validating decisions, highlighting the potential of LLM-based multi-agent systems to automate and enhance pest management processes.
Over the past year, the development of large language models (LLMs) has brought spatial intelligence into focus, with much attention on vision-based embodied intelligence. However, spatial intelligence spans a broader range of disciplines and scales, from navigation and urban planning to remote sensing and earth science. What are the differences and connections between spatial intelligence across these fields? In this paper, we first review human spatial cognition and its implications for spatial intelligence in LLMs. We then examine spatial memory, knowledge representations, and abstract reasoning in LLMs, highlighting their roles and connections. Finally, we analyze spatial intelligence across scales -- from embodied to urban and global levels -- following a framework that progresses from spatial memory and understanding to spatial reasoning and intelligence. Through this survey, we aim to provide insights into interdisciplinary spatial intelligence research and inspire future studies.
The proliferation of autonomous agents powered by large language models (LLMs) has revolutionized popular business applications dealing with tabular data, i.e., tabular agents. Although LLMs are observed to be vulnerable against prompt injection attacks from external data sources, tabular agents impose strict data formats and predefined rules on the attacker's payload, which are ineffective unless the agent navigates multiple layers of structural data to incorporate the payload. To address the challenge, we present a novel attack termed StruPhantom which specifically targets black-box LLM-powered tabular agents. Our attack designs an evolutionary optimization procedure which continually refines attack payloads via the proposed constrained Monte Carlo Tree Search augmented by an off-topic evaluator. StruPhantom helps systematically explore and exploit the weaknesses of target applications to achieve goal hijacking. Our evaluation validates the effectiveness of StruPhantom across various LLM-based agents, including those on real-world platforms, and attack scenarios. Our attack achieves over 50% higher success rates than baselines in enforcing the application's response to contain phishing links or malicious codes.
The reasoning capabilities of large language models (LLMs), such as OpenAI's o1 and DeepSeek-R1, have seen substantial advancements through deep thinking. However, these enhancements come with significant resource demands, underscoring the need to explore strategies to train effective reasoning LLMs with far fewer parameters. A critical challenge is that smaller models have different capacities and cognitive trajectories than their larger counterparts. Hence, direct distillation of chain-of-thought (CoT) results from large LLMs to smaller ones can be sometimes ineffective and requires a huge amount of annotated data. In this paper, we introduce a novel framework called Critique-Rethink-Verify (CRV), designed for training smaller yet powerful reasoning LLMs. Our CRV framework consists of multiple LLM agents, each specializing in unique abilities: (i) critiquing the CoTs according to the cognitive capabilities of smaller models, (ii) rethinking and refining these CoTs based on the critiques, and (iii) verifying the correctness of the refined results. We further propose the cognitive preference optimization (CogPO) algorithm to enhance the reasoning abilities of smaller models by aligning thoughts of these models with their cognitive capacities. Comprehensive evaluations on challenging reasoning benchmarks demonstrate the efficacy of CRV and CogPO, which outperforms other training methods by a large margin.
While large language models (LLMs) have demonstrated strong capabilities in tasks like question answering and fact verification, they continue to suffer from hallucinations and reasoning errors, especially in multi-hop tasks that require integration of multiple information sources. Current methods address these issues through retrieval-based techniques (grounding reasoning in external evidence), reasoning-based approaches (enhancing coherence via improved prompting), or hybrid strategies combining both elements. One prominent hybrid method, ReAct, has outperformed purely retrieval-based or reasoning-based approaches; however, it lacks internal verification of intermediate reasoning steps, allowing potential errors to propagate through complex reasoning tasks. In this paper, we introduce Reasoning Court (RC), a novel framework that extends iterative reasoning-and-retrieval methods, such as ReAct, with a dedicated LLM judge. Unlike ReAct, RC employs this judge to independently evaluate multiple candidate answers and their associated reasoning generated by separate LLM agents. The judge is asked to select the answer that it considers the most factually grounded and logically coherent based on the presented reasoning and evidence, or synthesizes a new answer using available evidence and its pre-trained knowledge if all candidates are inadequate, flawed, or invalid. Evaluations on multi-hop benchmarks (HotpotQA, MuSiQue) and fact-verification (FEVER) demonstrate that RC consistently outperforms state-of-the-art few-shot prompting methods without task-specific fine-tuning.
Multi-agent systems (MAS) built on large language models (LLMs) offer a promising path toward solving complex, real-world tasks that single-agent systems often struggle to manage. While recent advancements in test-time scaling (TTS) have significantly improved single-agent performance on challenging reasoning tasks, how to effectively scale collaboration and reasoning in MAS remains an open question. In this work, we introduce an adaptive multi-agent framework designed to enhance collaborative reasoning through both model-level training and system-level coordination. We construct M500, a high-quality dataset containing 500 multi-agent collaborative reasoning traces, and fine-tune Qwen2.5-32B-Instruct on this dataset to produce M1-32B, a model optimized for multi-agent collaboration. To further enable adaptive reasoning, we propose a novel CEO agent that dynamically manages the discussion process, guiding agent collaboration and adjusting reasoning depth for more effective problem-solving. Evaluated in an open-source MAS across a range of tasks-including general understanding, mathematical reasoning, and coding-our system significantly outperforms strong baselines. For instance, M1-32B achieves 12% improvement on GPQA-Diamond, 41% on AIME2024, and 10% on MBPP-Sanitized, matching the performance of state-of-the-art models like DeepSeek-R1 on some tasks. These results highlight the importance of both learned collaboration and adaptive coordination in scaling multi-agent reasoning. Code is available at https://github.com/jincan333/MAS-TTS
Peer review at AI conferences is stressed by rapidly rising submission volumes, leading to deteriorating review quality and increased author dissatisfaction. To address these issues, we developed Review Feedback Agent, a system leveraging multiple large language models (LLMs) to improve review clarity and actionability by providing automated feedback on vague comments, content misunderstandings, and unprofessional remarks to reviewers. Implemented at ICLR 2025 as a large randomized control study, our system provided optional feedback to more than 20,000 randomly selected reviews. To ensure high-quality feedback for reviewers at this scale, we also developed a suite of automated reliability tests powered by LLMs that acted as guardrails to ensure feedback quality, with feedback only being sent to reviewers if it passed all the tests. The results show that 27% of reviewers who received feedback updated their reviews, and over 12,000 feedback suggestions from the agent were incorporated by those reviewers. This suggests that many reviewers found the AI-generated feedback sufficiently helpful to merit updating their reviews. Incorporating AI feedback led to significantly longer reviews (an average increase of 80 words among those who updated after receiving feedback) and more informative reviews, as evaluated by blinded researchers. Moreover, reviewers who were selected to receive AI feedback were also more engaged during paper rebuttals, as seen in longer author-reviewer discussions. This work demonstrates that carefully designed LLM-generated review feedback can enhance peer review quality by making reviews more specific and actionable while increasing engagement between reviewers and authors. The Review Feedback Agent is publicly available at https://github.com/zou-group/review_feedback_agent.
This paper introduces a methodology based on agentic workflows for economic research that leverages Large Language Models (LLMs) and multimodal AI to enhance research efficiency and reproducibility. Our approach features autonomous and iterative processes covering the entire research lifecycle--from ideation and literature review to economic modeling and data processing, empirical analysis and result interpretation--with strategic human oversight. The workflow architecture comprises specialized agents with clearly defined roles, structured inter-agent communication protocols, systematic error escalation pathways, and adaptive mechanisms that respond to changing research demand. Human-in-the-loop (HITL) checkpoints are strategically integrated to ensure methodological validity and ethical compliance. We demonstrate the practical implementation of our framework using Microsoft's open-source platform, AutoGen, presenting experimental examples that highlight both the current capabilities and future potential of agentic workflows in improving economic research.
A/B testing experiment is a widely adopted method for evaluating UI/UX design decisions in modern web applications. Yet, traditional A/B testing remains constrained by its dependence on the large-scale and live traffic of human participants, and the long time of waiting for the testing result. Through formative interviews with six experienced industry practitioners, we identified critical bottlenecks in current A/B testing workflows. In response, we present AgentA/B, a novel system that leverages Large Language Model-based autonomous agents (LLM Agents) to automatically simulate user interaction behaviors with real webpages. AgentA/B enables scalable deployment of LLM agents with diverse personas, each capable of navigating the dynamic webpage and interactively executing multi-step interactions like search, clicking, filtering, and purchasing. In a demonstrative controlled experiment, we employ AgentA/B to simulate a between-subject A/B testing with 1,000 LLM agents Amazon.com, and compare agent behaviors with real human shopping behaviors at a scale. Our findings suggest AgentA/B can emulate human-like behavior patterns.
Existing evaluation of large language model (LLM) agents on scientific discovery lacks objective baselines and metrics to assess the viability of their proposed methods. To address this issue, we introduce MLRC-Bench, a benchmark designed to quantify how effectively language agents can tackle challenging Machine Learning (ML) Research Competitions. Our benchmark highlights open research problems that demand novel methodologies, in contrast to recent benchmarks such as OpenAI's MLE-Bench (Chan et al., 2024) and METR's RE-Bench (Wijk et al., 2024), which focus on well-established research tasks that are largely solvable through sufficient engineering effort. Unlike prior work, e.g., AI Scientist (Lu et al., 2024b), which evaluates the end-to-end agentic pipeline by using LLM-as-a-judge, MLRC-Bench measures the key steps of proposing and implementing novel research methods and evaluates them with newly proposed rigorous protocol and objective metrics. Our curated suite of 7 competition tasks reveals significant challenges for LLM agents. Even the best-performing tested agent (gemini-exp-1206 under MLAB (Huang et al., 2024a)) closes only 9.3% of the gap between baseline and top human participant scores. Furthermore, our analysis reveals a misalignment between the LLM-judged innovation and their actual performance on cutting-edge ML research problems. MLRC-Bench is a dynamic benchmark, which is designed to continually grow with new ML competitions to encourage rigorous and objective evaluations of AI's research capabilities.
The rise of LLM-driven AI characters raises safety concerns, particularly for vulnerable human users with psychological disorders. To address these risks, we propose EmoAgent, a multi-agent AI framework designed to evaluate and mitigate mental health hazards in human-AI interactions. EmoAgent comprises two components: EmoEval simulates virtual users, including those portraying mentally vulnerable individuals, to assess mental health changes before and after interactions with AI characters. It uses clinically proven psychological and psychiatric assessment tools (PHQ-9, PDI, PANSS) to evaluate mental risks induced by LLM. EmoGuard serves as an intermediary, monitoring users' mental status, predicting potential harm, and providing corrective feedback to mitigate risks. Experiments conducted in popular character-based chatbots show that emotionally engaging dialogues can lead to psychological deterioration in vulnerable users, with mental state deterioration in more than 34.4% of the simulations. EmoGuard significantly reduces these deterioration rates, underscoring its role in ensuring safer AI-human interactions. Our code is available at: https://github.com/1akaman/EmoAgent