We introduce DeepResearch$^{\text{Eco}}$, a novel agentic LLM-based system for automated scientific synthesis that supports recursive, depth- and breadth-controlled exploration of original research questions -- enhancing search diversity and nuance in the retrieval of relevant scientific literature. Unlike conventional retrieval-augmented generation pipelines, DeepResearch enables user-controllable synthesis with transparent reasoning and parameter-driven configurability, facilitating high-throughput integration of domain-specific evidence while maintaining analytical rigor. Applied to 49 ecological research questions, DeepResearch achieves up to a 21-fold increase in source integration and a 14.9-fold rise in sources integrated per 1,000 words. High-parameter settings yield expert-level analytical depth and contextual diversity. Source code available at: https://github.com/sciknoworg/deep-research.
The integration of large language models (LLMs) into enterprise systems has created a new class of covert security vulnerabilities, particularly within logic-execution layers and persistent-memory contexts. In this paper, we introduce Logic-Layer Prompt Control Injection (LPCI), a novel attack category in which encoded, delayed, and conditionally triggered payloads are embedded in memory, vector stores, or tool outputs. These payloads can bypass conventional input filters and trigger unauthorised behaviour across sessions.
Visual coverage path planning with unmanned aerial vehicles (UAVs) requires agents to strategically coordinate UAV motion and camera control to maximize coverage, minimize redundancy, and maintain battery efficiency. Traditional reinforcement learning (RL) methods rely on environment-specific reward formulations that lack semantic adaptability. This study proposes Prompt-Informed Reinforcement Learning (PIRL), a novel approach that integrates the zero-shot reasoning ability and in-context learning capability of large language models with curiosity-driven RL. PIRL leverages semantic feedback from an LLM, GPT-3.5, to dynamically shape the reward function of the Proximal Policy Optimization (PPO) RL policy guiding the agent in position and camera adjustments for optimal visual coverage. The PIRL agent is trained using OpenAI Gym and evaluated in various environments. Furthermore, the sim-to-real-like ability and zero-shot generalization of the agent are tested by operating the agent in Webots simulator which introduces realistic physical dynamics. Results show that PIRL outperforms multiple learning-based baselines such as PPO with static rewards, PPO with exploratory weight initialization, imitation learning, and an LLM-only controller. Across different environments, PIRL outperforms the best-performing baseline by achieving up to 14% higher visual coverage in OpenAI Gym and 27% higher in Webots, up to 25% higher battery efficiency, and up to 18\% lower redundancy, depending on the environment. The results highlight the effectiveness of LLM-guided reward shaping in complex spatial exploration tasks and suggest a promising direction for integrating natural language priors into RL for robotics.
Tables are fundamental in domains such as finance, healthcare, and public administration, yet real-world table tasks often involve noise, structural heterogeneity, and semantic complexity--issues underexplored in existing research that primarily targets clean academic datasets. This survey focuses on LLM-based Table Agents, which aim to automate table-centric workflows by integrating preprocessing, reasoning, and domain adaptation. We define five core competencies--C1: Table Structure Understanding, C2: Table and Query Semantic Understanding, C3: Table Retrieval and Compression, C4: Executable Reasoning with Traceability, and C5: Cross-Domain Generalization--to analyze and compare current approaches. In addition, a detailed examination of the Text-to-SQL Agent reveals a performance gap between academic benchmarks and real-world scenarios, especially for open-source models. Finally, we provide actionable insights to improve the robustness, generalization, and efficiency of LLM-based Table Agents in practical settings.
Are AI systems truly representing human values, or merely averaging across them? Our study suggests a concerning reality: Large Language Models (LLMs) fail to represent diverse cultural moral frameworks despite their linguistic capabilities. We expose significant gaps between AI-generated and human moral intuitions by applying the Moral Foundations Questionnaire across 19 cultural contexts. Comparing multiple state-of-the-art LLMs' origins against human baseline data, we find these models systematically homogenize moral diversity. Surprisingly, increased model size doesn't consistently improve cultural representation fidelity. Our findings challenge the growing use of LLMs as synthetic populations in social science research and highlight a fundamental limitation in current AI alignment approaches. Without data-driven alignment beyond prompting, these systems cannot capture the nuanced, culturally-specific moral intuitions. Our results call for more grounded alignment objectives and evaluation metrics to ensure AI systems represent diverse human values rather than flattening the moral landscape.
Our research uncovers a novel privacy risk associated with multimodal large language models (MLLMs): the ability to infer sensitive personal attributes from audio data -- a technique we term audio private attribute profiling. This capability poses a significant threat, as audio can be covertly captured without direct interaction or visibility. Moreover, compared to images and text, audio carries unique characteristics, such as tone and pitch, which can be exploited for more detailed profiling. However, two key challenges exist in understanding MLLM-employed private attribute profiling from audio: (1) the lack of audio benchmark datasets with sensitive attribute annotations and (2) the limited ability of current MLLMs to infer such attributes directly from audio. To address these challenges, we introduce AP^2, an audio benchmark dataset that consists of two subsets collected and composed from real-world data, and both are annotated with sensitive attribute labels. Additionally, we propose Gifts, a hybrid multi-agent framework that leverages the complementary strengths of audio-language models (ALMs) and large language models (LLMs) to enhance inference capabilities. Gifts employs an LLM to guide the ALM in inferring sensitive attributes, then forensically analyzes and consolidates the ALM's inferences, overcoming severe hallucinations of existing ALMs in generating long-context responses. Our evaluations demonstrate that Gifts significantly outperforms baseline approaches in inferring sensitive attributes. Finally, we investigate model-level and data-level defense strategies to mitigate the risks of audio private attribute profiling. Our work validates the feasibility of audio-based privacy attacks using MLLMs, highlighting the need for robust defenses, and provides a dataset and framework to facilitate future research.
Recent advances in Large Language Models (LLM) have led to a new class of autonomous agents, renewing and expanding interest in the area. LLM-powered Multiagent Systems (MAS) have thus emerged, both for assistive and simulation purposes, yet tools for realistic human behavior simulation -- with its distinctive challenges and opportunities -- remain underdeveloped. Existing MAS libraries and tools lack fine-grained persona specifications, population sampling facilities, experimentation support, and integrated validation, among other key capabilities, limiting their utility for behavioral studies, social simulation, and related applications. To address these deficiencies, in this work we introduce TinyTroupe, a simulation toolkit enabling detailed persona definitions (e.g., nationality, age, occupation, personality, beliefs, behaviors) and programmatic control via numerous LLM-driven mechanisms. This allows for the concise formulation of behavioral problems of practical interest, either at the individual or group level, and provides effective means for their solution. TinyTroupe's components are presented using representative working examples, such as brainstorming and market research sessions, thereby simultaneously clarifying their purpose and demonstrating their usefulness. Quantitative and qualitative evaluations of selected aspects are also provided, highlighting possibilities, limitations, and trade-offs. The approach, though realized as a specific Python implementation, is meant as a novel conceptual contribution, which can be partially or fully incorporated in other contexts. The library is available as open source at https://github.com/microsoft/tinytroupe.
We use generative agents powered by large language models (LLMs) to simulate a social network in a shared residential building, driving the temperature decisions for a central heating system. Agents, divided into Family Members and Representatives, consider personal preferences, personal traits, connections, and weather conditions. Daily simulations involve family-level consensus followed by building-wide decisions among representatives. We tested three personality traits distributions (positive, mixed, and negative) and found that positive traits correlate with higher happiness and stronger friendships. Temperature preferences, assertiveness, and selflessness have a significant impact on happiness and decisions. This work demonstrates how LLM-driven agents can help simulate nuanced human behavior where complex real-life human simulations are difficult to set.
We introduce the THOR (Transformer Heuristics for On-Demand Retrieval) Module, designed and implemented by eSapiens, a secure, scalable engine that transforms natural-language questions into verified, read-only SQL analytics for enterprise databases. The Text-to-SQL module follows a decoupled orchestration/execution architecture: a Supervisor Agent routes queries, Schema Retrieval dynamically injects table and column metadata, and a SQL Generation Agent emits single-statement SELECT queries protected by a read-only guardrail. An integrated Self-Correction & Rating loop captures empty results, execution errors, or low-quality outputs and triggers up to five LLM-driven regeneration attempts. Finally, a Result Interpretation Agent produces concise, human-readable insights and hands raw rows to the Insight & Intelligence engine for visualization or forecasting. Smoke tests across finance, sales, and operations scenarios demonstrate reliable ad-hoc querying and automated periodic reporting. By embedding schema awareness, fault-tolerant execution, and compliance guardrails, the THOR Module empowers non-technical users to access live data with zero-SQL simplicity and enterprise-grade safety.
We present eSapiens, an AI-as-a-Service (AIaaS) platform engineered around a business-oriented trifecta: proprietary data, operational workflows, and any major agnostic Large Language Model (LLM). eSapiens gives businesses full control over their AI assets, keeping everything in-house for AI knowledge retention and data security. eSapiens AI Agents (Sapiens) empower your team by providing valuable insights and automating repetitive tasks, enabling them to focus on high-impact work and drive better business outcomes. The system integrates structured document ingestion, hybrid vector retrieval, and no-code orchestration via LangChain, and supports top LLMs including OpenAI, Claude, Gemini, and DeepSeek. A key component is the THOR Agent, which handles structured SQL-style queries and generates actionable insights over enterprise databases. To evaluate the system, we conduct two experiments. First, a retrieval benchmark on legal corpora reveals that a chunk size of 512 tokens yields the highest retrieval precision (Top-3 accuracy: 91.3%). Second, a generation quality test using TRACe metrics across five LLMs shows that eSapiens delivers more context-consistent outputs with up to a 23% improvement in factual alignment. These results demonstrate the effectiveness of eSapiens in enabling trustworthy, auditable AI workflows for high-stakes domains like legal and finance.
Large language models (LLMs) have demonstrated remarkable capabilities across a variety of domains. However, their applications in cryptography, which serves as a foundational pillar of cybersecurity, remain largely unexplored. To address this gap, we propose \textbf{AICrypto}, the first comprehensive benchmark designed to evaluate the cryptographic capabilities of LLMs. The benchmark comprises 135 multiple-choice questions, 150 capture-the-flag (CTF) challenges, and 18 proof problems, covering a broad range of skills from factual memorization to vulnerability exploitation and formal reasoning. All tasks are carefully reviewed or constructed by cryptography experts to ensure correctness and rigor. To support automated evaluation of CTF challenges, we design an agent-based framework. To gain deeper insight into the current state of cryptographic proficiency in LLMs, we introduce human expert performance baselines for comparison across all task types. Our evaluation of 17 leading LLMs reveals that state-of-the-art models match or even surpass human experts in memorizing cryptographic concepts, exploiting common vulnerabilities, and routine proofs. However, they still lack a deep understanding of abstract mathematical concepts and struggle with tasks that require multi-step reasoning and dynamic analysis. We hope this work could provide insights for future research on LLMs in cryptographic applications. Our code and dataset are available at https://aicryptobench.github.io.
By integrating tools from external APIs, Large Language Models (LLMs) have expanded their promising capabilities in a diverse spectrum of complex real-world tasks. However, testing, evaluation, and analysis of LLM tool use remain in their early stages. Most existing benchmarks rely on manually collected test cases, many of which cannot be automatically checked for semantic correctness and instead depend on static methods such as string matching. Additionally, these benchmarks often overlook the complex interactions that occur between sequential API calls, which are common in real-world applications. To fill the gap, in this paper, we introduce StateGen, an automated framework designed to generate diverse coding tasks involving sequential API interactions. StateGen combines state-machine-based API constraint solving and validation, energy-based sampling, and control-flow injection to generate executable programs. These programs are then translated into human-like natural language task descriptions through a collaboration of two LLM agents. Utilizing StateGen, we construct StateEval, a benchmark encompassing 120 verified test cases spanning across three representative scenarios: Session Service, Tensor Operation, and ElevenLabs MCP. Experimental results confirm that StateGen can effectively generate challenging and realistic API-oriented tasks, highlighting areas for improvement in current LLMs incorporating APIs.
Retrieval-Augmented Generation (RAG) lifts the factuality of Large Language Models (LLMs) by injecting external knowledge, yet it falls short on problems that demand multi-step inference; conversely, purely reasoning-oriented approaches often hallucinate or mis-ground facts. This survey synthesizes both strands under a unified reasoning-retrieval perspective. We first map how advanced reasoning optimizes each stage of RAG (Reasoning-Enhanced RAG). Then, we show how retrieved knowledge of different type supply missing premises and expand context for complex inference (RAG-Enhanced Reasoning). Finally, we spotlight emerging Synergized RAG-Reasoning frameworks, where (agentic) LLMs iteratively interleave search and reasoning to achieve state-of-the-art performance across knowledge-intensive benchmarks. We categorize methods, datasets, and open challenges, and outline research avenues toward deeper RAG-Reasoning systems that are more effective, multimodally-adaptive, trustworthy, and human-centric. The collection is available at https://github.com/DavidZWZ/Awesome-RAG-Reasoning.
We introduce the framework of LLM-Stackelberg games, a class of sequential decision-making models that integrate large language models (LLMs) into strategic interactions between a leader and a follower. Departing from classical Stackelberg assumptions of complete information and rational agents, our formulation allows each agent to reason through structured prompts, generate probabilistic behaviors via LLMs, and adapt their strategies through internal cognition and belief updates. We define two equilibrium concepts: reasoning and behavioral equilibrium, which aligns an agent's internal prompt-based reasoning with observable behavior, and conjectural reasoning equilibrium, which accounts for epistemic uncertainty through parameterized models over an opponent's response. These layered constructs capture bounded rationality, asymmetric information, and meta-cognitive adaptation. We illustrate the framework through a spearphishing case study, where a sender and a recipient engage in a deception game using structured reasoning prompts. This example highlights the cognitive richness and adversarial potential of LLM-mediated interactions. Our results show that LLM-Stackelberg games provide a powerful paradigm for modeling decision-making in domains such as cybersecurity, misinformation, and recommendation systems.
Explainability and interpretability are cornerstones of frontier and next-generation artificial intelligence (AI) systems. This is especially true in recent systems, such as large language models (LLMs), and more broadly, generative AI. On the other hand, adaptability to new domains, contexts, or scenarios is also an important aspect for a successful system. As such, we are particularly interested in how we can merge these two efforts, that is, investigating the design of transferable and interpretable neurosymbolic AI systems. Specifically, we focus on a class of systems referred to as ''Agentic Retrieval-Augmented Generation'' systems, which actively select, interpret, and query knowledge sources in response to natural language prompts. In this paper, we systematically evaluate how different conceptualizations and representations of knowledge, particularly the structure and complexity, impact an AI agent (in this case, an LLM) in effectively querying a triplestore. We report our results, which show that there are impacts from both approaches, and we discuss their impact and implications.
LLM-based coding agents are rapidly being deployed in software development, yet their security implications remain poorly understood. These agents, while capable of accelerating software development, may inadvertently introduce insecure practices. We conducted the first systematic security evaluation of autonomous coding agents, analyzing over 12,000 actions across five state-of-the-art models (GPT-4o, GPT-4.1, Claude variants) on 93 real-world software setup tasks. Our findings reveal significant security concerns: 21% of agent trajectories contained insecure actions, with models showing substantial variation in security behavior. We developed a high-precision detection system that identified four major vulnerability categories, with information exposure (CWE-200) being the most prevalent one. We also evaluated mitigation strategies including feedback mechanisms and security reminders with various effectiveness between models. GPT-4.1 demonstrated exceptional security awareness with 96.8% mitigation success. Our work provides the first comprehensive framework for evaluating coding agent security and highlights the need for security-aware design of next generation LLM-based coding agents.
We present StockSim, an open-source simulation platform for systematic evaluation of large language models (LLMs) in realistic financial decision-making scenarios. Unlike previous toolkits that offer limited scope, StockSim delivers a comprehensive system that fully models market dynamics and supports diverse simulation modes of varying granularity. It incorporates critical real-world factors, such as latency, slippage, and order-book microstructure, that were previously neglected, enabling more faithful and insightful assessment of LLM-based trading agents. An extensible, role-based agent framework supports heterogeneous trading strategies and multi-agent coordination, making StockSim a uniquely capable testbed for NLP research on reasoning under uncertainty and sequential decision-making. We open-source all our code at https: //github.com/harrypapa2002/StockSim.
Decentralized finance (DeFi) has introduced a new era of permissionless financial innovation but also led to unprecedented market manipulation. Without centralized oversight, malicious actors coordinate shilling campaigns and pump-and-dump schemes across various platforms. We propose a Multi-Agent Reinforcement Learning (MARL) framework for decentralized manipulation detection, modeling the interaction between manipulators and detectors as a dynamic adversarial game. This framework identifies suspicious patterns using delayed token price reactions as financial indicators.Our method introduces three innovations: (1) Group Relative Policy Optimization (GRPO) to enhance learning stability in sparse-reward and partially observable settings; (2) a theory-based reward function inspired by rational expectations and information asymmetry, differentiating price discovery from manipulation noise; and (3) a multi-modal agent pipeline that integrates LLM-based semantic features, social graph signals, and on-chain market data for informed decision-making.The framework is integrated within the Symphony system, a decentralized multi-agent architecture enabling peer-to-peer agent execution and trust-aware learning through distributed logs, supporting chain-verifiable evaluation. Symphony promotes adversarial co-evolution among strategic actors and maintains robust manipulation detection without centralized oracles, enabling real-time surveillance across global DeFi ecosystems.Trained on 100,000 real-world discourse episodes and validated in adversarial simulations, Hide-and-Shill achieves top performance in detection accuracy and causal attribution. This work bridges multi-agent systems with financial surveillance, advancing a new paradigm for decentralized market intelligence. All resources are available at the Hide-and-Shill GitHub repository to promote open research and reproducibility.
In decision-making conversations, experts must navigate complex choices and make on-the-spot decisions while engaged in conversation. Although extensive historical data often exists, the real-time nature of these scenarios makes it infeasible for decision-makers to review and leverage relevant information. This raises an interesting question: What if experts could utilize relevant past data in real-time decision-making through insights derived from past data? To explore this, we implemented a conversational user interface, taking doctor-patient interactions as an example use case. Our system continuously listens to the conversation, identifies patient problems and doctor-suggested solutions, and retrieves related data from an embedded dataset, generating concise insights using a pipeline built around a retrieval-based Large Language Model (LLM) agent. We evaluated the prototype by embedding Health Canada datasets into a vector database and conducting simulated studies using sample doctor-patient dialogues, showing effectiveness but also challenges, setting directions for the next steps of our work.
This paper investigates the behavior of simulated AI agents (large language models, or LLMs) in auctions, introducing a novel synthetic data-generating process to help facilitate the study and design of auctions. We find that LLMs -- when endowed with chain of thought reasoning capacity -- agree with the experimental literature in auctions across a variety of classic auction formats. In particular, we find that LLM bidders produce results consistent with risk-averse human bidders; that they perform closer to theoretical predictions in obviously strategy-proof auctions; and, that they succumb to the winner's curse in common value settings. On prompting, we find that LLMs are not very sensitive to naive changes in prompts (e.g., language, currency) but can improve dramatically towards theoretical predictions with the right mental model (i.e., the language of Nash deviations). We run 1,000$+$ auctions for less than $\$$400 with GPT-4 models (three orders of magnitude cheaper than modern auction experiments) and develop a framework flexible enough to run auction experiments with any LLM model and a wide range of auction design specifications, facilitating further experimental study by decreasing costs and serving as a proof-of-concept for the use of LLM proxies.
The rapid advancements in Large Language Models (LLMs) and their multimodal extensions (MLLMs) have ushered in remarkable progress in video understanding. However, a fundamental challenge persists: effectively processing and comprehending video content that extends beyond minutes or hours. While recent efforts like Video-XL-2 have demonstrated novel architectural solutions for extreme efficiency, and advancements in positional encoding such as HoPE and VideoRoPE++ aim to improve spatio-temporal understanding over extensive contexts, current state-of-the-art models still encounter significant computational and memory constraints when faced with the sheer volume of visual tokens from lengthy sequences. Furthermore, maintaining temporal coherence, tracking complex events, and preserving fine-grained details over extended periods remain formidable hurdles, despite progress in agentic reasoning systems like Deep Video Discovery. This position paper posits that a logical, albeit ambitious, next frontier for multimedia research is Infinite Video Understanding -- the capability for models to continuously process, understand, and reason about video data of arbitrary, potentially never-ending duration. We argue that framing Infinite Video Understanding as a blue-sky research objective provides a vital north star for the multimedia, and the wider AI, research communities, driving innovation in areas such as streaming architectures, persistent memory mechanisms, hierarchical and adaptive representations, event-centric reasoning, and novel evaluation paradigms. Drawing inspiration from recent work on long/ultra-long video understanding and several closely related fields, we outline the core challenges and key research directions towards achieving this transformative capability.
Modern Large Language Model (LLM) agents promise end to end assistance with real-world software tasks, yet existing benchmarks evaluate LLM agents almost exclusively in pre-baked environments where every dependency is pre-installed. To fill this gap, we introduce SetupBench, a 93 instance benchmark that isolates the environment-bootstrap skill: starting from a bare Linux sandbox, an agent must install packages, resolve dependency conflicts, initialize databases, and configure background services. Our tasks span seven language ecosystems, five database engines, and multi-service orchestration scenarios, each accompanies by a natural language problem statement and a deterministic success command. Through evaluation of OpenHands, a state-of-the-art coding agent, we find low success rates across task categories, with particular challenges in repository setup (38.9-57.4%) and local database configuration (20.0-53.3%). Our analysis reveals systematic failure modes including incomplete development tooling installation, hallucinated task constraints, and non-persistent environment modifications that break agent-human collaboration workflows. We identify substantial inefficiencies in agent exploration strategies, with 38-89% of actions being unnecessary compared to optimal human behavior. These findings highlight gaps in current agents' practical environment-bootstrap capabilities. By targeting this critical yet under-evaluated capability, SetupBench provides a rigorous yard-stick for the next generation of software developer agents aiming to solve end to end real-wold tasks.
Large Language Models (LLMs) have achieved strong performance on a wide range of complex reasoning tasks, yet further gains are often possible by leveraging the complementary strengths of multiple models. While multi-agent frameworks can improve solution quality by leveraging multiple LLMs, existing methods are often computationally expensive, both at training and inference time. In this work, we introduce a hierarchical multi-agent framework that addresses these challenges by training only a single leader LLM to coordinate a team of untrained peer agents. To this end, we propose Multi-agent guided Leader Policy \textbf{O}ptimization (MLPO), a novel approach which trains the leader to evaluate and synthesize agent responses without auxiliary value networks or explicit agent feedback. Leaders trained with MLPO exhibit improved performance not only when interacting with the agent team at inference time, but also enjoy improved performance when deployed in single-agent settings without the team. Empirical results on Big-Bench Hard (BBH), MATH, and MMLU demonstrate that our framework achieves substantial performance improvements over both single-agent and multi-agent baselines. Our results highlight the effectiveness and efficiency of training a single, flexible leader for collaborative reasoning in multi-agent LLM systems.
As cosmological simulations and their associated software become increasingly complex, physicists face the challenge of searching through vast amounts of literature and user manuals to extract simulation parameters from dense academic papers, each using different models and formats. Translating these parameters into executable scripts remains a time-consuming and error-prone process. To improve efficiency in physics research and accelerate the cosmological simulation process, we introduce SimAgents, a multi-agent system designed to automate both parameter configuration from the literature and preliminary analysis for cosmology research. SimAgents is powered by specialized LLM agents capable of physics reasoning, simulation software validation, and tool execution. These agents collaborate through structured communication, ensuring that extracted parameters are physically meaningful, internally consistent, and software-compliant. We also construct a cosmological parameter extraction evaluation dataset by collecting over 40 simulations in published papers from Arxiv and leading journals that cover diverse simulation types. Experiments on the dataset demonstrate a strong performance of SimAgents, highlighting its effectiveness and potential to accelerate scientific research for physicists. Our demonstration video is available at: https://youtu.be/w1zLpm_CaWA. The complete system and dataset are publicly available at https://github.com/xwzhang98/SimAgents.
Large language model (LLM)-based multi-agent systems have demonstrated remarkable promise for tackling complex tasks by breaking them down into subtasks that are iteratively planned, executed, observed, and refined. Despite their effectiveness, these systems often incur high latency because real-world problems frequently demand multiple iterative cycles of reasoning steps. To address this challenge, we propose M1-Parallel, a framework that concurrently runs multiple multi-agent teams in parallel to uncover distinct solution paths. By leveraging an event-driven communication model with asynchronous messaging, M1-Parallel efficiently capitalizes on the inherent diversity of valid plans to either reduce end-to-end latency or boost task completion rates. Our experiments on complex tasks show that M1-Parallel with early termination achieves up to $2.2\times$ speedup while preserving accuracy, and that M1-Parallel with aggregation yields higher task completion rates. We further investigate strategies aimed at encouraging diverse execution plans but observe no additional performance gains over repeated sampling. Overall, these findings underscore the potential of parallel plan execution for optimizing multi-agent systems for real-world, high-complexity reasoning tasks.
We present elsciRL, an open-source Python library to facilitate the application of language solutions on reinforcement learning problems. We demonstrate the potential of our software by extending the Language Adapter with Self-Completing Instruction framework defined in (Osborne, 2024) with the use of LLMs. Our approach can be re-applied to new applications with minimal setup requirements. We provide a novel GUI that allows a user to provide text input for an LLM to generate instructions which it can then self-complete. Empirical results indicate that these instructions \textit{can} improve a reinforcement learning agent's performance. Therefore, we present this work to accelerate the evaluation of language solutions on reward based environments to enable new opportunities for scientific discovery.
AI Agents rely on Large Language Models (LLMs) and Multimodal-LLMs (MLLMs) to perform interpretation and inference in text and image tasks without post-training, where LLMs and MLLMs play the most critical role and determine the initial ability and limitations of AI Agents. Usually, AI Agents utilize sophisticated prompt engineering and external reasoning framework to obtain a promising interaction with LLMs, e.g., Chain-of-Thought, Iteration of Thought and Image-of-Thought. However, they are still constrained by the inherent limitations of LLM in understanding natural language, and the iterative reasoning process will generate a large amount of inference cost. To this end, we propose a novel AI Agent Reasoning Framework with Introspection of Thought (INoT) by designing a new LLM-Read code in prompt. It enables LLM to execute programmatic dialogue reasoning processes following the code in prompt. Therefore, self-denial and reflection occur within LLM instead of outside LLM, which can reduce token cost effectively. Through our experiments on six benchmarks for three different tasks, the effectiveness of INoT is verified, with an average improvement of 7.95\% in performance, exceeding the baselines. Furthermore, the token cost of INoT is lower on average than the best performing method at baseline by 58.3\%. In addition, we demonstrate the versatility of INoT in image interpretation and inference through verification experiments.
Early-stage engineering design involves complex, iterative reasoning, yet existing large language model (LLM) workflows struggle to maintain task continuity and generate executable models. We evaluate whether a structured multi-agent system (MAS) can more effectively manage requirements extraction, functional decomposition, and simulator code generation than a simpler two-agent system (2AS). The target application is a solar-powered water filtration system as described in a cahier des charges. We introduce the Design-State Graph (DSG), a JSON-serializable representation that bundles requirements, physical embodiments, and Python-based physics models into graph nodes. A nine-role MAS iteratively builds and refines the DSG, while the 2AS collapses the process to a Generator-Reflector loop. Both systems run a total of 60 experiments (2 LLMs - Llama 3.3 70B vs reasoning-distilled DeepSeek R1 70B x 2 agent configurations x 3 temperatures x 5 seeds). We report a JSON validity, requirement coverage, embodiment presence, code compatibility, workflow completion, runtime, and graph size. Across all runs, both MAS and 2AS maintained perfect JSON integrity and embodiment tagging. Requirement coverage remained minimal (less than 20\%). Code compatibility peaked at 100\% under specific 2AS settings but averaged below 50\% for MAS. Only the reasoning-distilled model reliably flagged workflow completion. Powered by DeepSeek R1 70B, the MAS generated more granular DSGs (average 5-6 nodes) whereas 2AS mode-collapsed. Structured multi-agent orchestration enhanced design detail. Reasoning-distilled LLM improved completion rates, yet low requirements and fidelity gaps in coding persisted.
Large-language models (LLMs) have demonstrated powerful problem-solving capabilities, in particular when organized in multi-agent systems. However, the advent of such systems also raises several questions on the ability of a complex network of agents to effectively self-organize and collaborate. While measuring performance on standard reasoning benchmarks indicates how well multi-agent systems can solve reasoning tasks, it is unclear whether these systems are able to leverage their topology effectively. Here, we propose AgentsNet, a new benchmark for multi-agent reasoning. By drawing inspiration from classical problems in distributed systems and graph theory, AgentsNet measures the ability of multi-agent systems to collaboratively form strategies for problem-solving, self-organization, and effective communication given a network topology. We evaluate a variety of baseline methods on AgentsNet including homogeneous networks of agents which first have to agree on basic protocols for organization and communication. We find that some frontier LLMs are already demonstrating strong performance for small networks but begin to fall off once the size of the network scales. While existing multi-agent benchmarks cover at most 2-5 agents, AgentsNet is practically unlimited in size and can scale with new generations of LLMs. As such, we also probe frontier models in a setup with up to 100 agents.
Image captioning is an important problem in developing various AI systems, and these tasks require large volumes of annotated images to train the models. Since all existing labelled datasets are already used for training the large Vision Language Models (VLMs), it becomes challenging to improve the performance of the same. Considering this, it is essential to consider the unsupervised image captioning performance, which remains relatively under-explored. To that end, we propose LoGIC (Lewis Communication Game for Image Captioning), a Multi-agent Reinforcement Learning game. The proposed method consists of two agents, a 'speaker' and a 'listener', with the objective of learning a strategy for communicating in natural language. We train agents in the cooperative common-reward setting using the GRPO algorithm and show that improvement in image captioning performance emerges as a consequence of the agents learning to play the game. We show that using pre-trained VLMs as the 'speaker' and Large Language Model (LLM) for language understanding in the 'listener', we achieved a $46$ BLEU score after fine-tuning using LoGIC without additional labels, a $2$ units advantage in absolute metrics compared to the $44$ BLEU score of the vanilla VLM. Additionally, we replace the VLM from the 'speaker' with lightweight components: (i) a ViT for image perception and (ii) a GPT2 language generation, and train them from scratch using LoGIC, obtaining a $31$ BLEU score in the unsupervised setting, a $10$ points advantage over existing unsupervised image-captioning methods.