Many state-of-the-art LLMs are trained to think before giving their answer. Reasoning can greatly improve language model capabilities and safety, but it also makes them less interactive: given a new input, a model must stop thinking before it can respond. Real-world use cases such as voice-based or embedded assistants require an LLM agent to respond and adapt to additional information in real time, which is incompatible with sequential interactions. In contrast, humans can listen, think, and act asynchronously: we begin thinking about the problem while reading it and continue thinking while formulating the answer. In this work, we augment LLMs capable of reasoning to operate in a similar way without additional training. Our method uses the properties of rotary embeddings to enable LLMs built for sequential interactions to simultaneously think, listen, and generate outputs. We evaluate our approach on math, commonsense, and safety reasoning and find that it can generate accurate thinking-augmented answers in real time, reducing time to first non-thinking token from minutes to <= 5s. and the overall real-time delays by 6-11x.
Social presence is central to the enjoyment of watching content together, yet modern media consumption is increasingly solitary. We investigate whether multi-agent conversational AI systems can recreate the dynamics of shared viewing experiences across diverse content types. We present CompanionCast, a general framework for orchestrating multiple role-specialized AI agents that respond to video content using multimodal inputs, speech synthesis, and spatial audio. Distinctly, CompanionCast integrates an LLM-as-a-Judge module that iteratively scores and refines conversations across five dimensions (relevance, authenticity, engagement, diversity, personality consistency). We validate this framework through sports viewing, a domain with rich dynamics and strong social traditions, where a pilot study with soccer fans suggests that multi-agent interaction improves perceived social presence compared to solo viewing. We contribute: (1) a generalizable framework for orchestrating multi-agent conversations around multimodal video content, (2) a novel evaluator-agent pipeline for conversation quality control, and (3) exploratory evidence of increased social presence in AI-mediated co-viewing. We discuss challenges and future directions for applying this approach to diverse viewing contexts including entertainment, education, and collaborative watching experiences.
Large language models (LLMs) have achieved significant progress in solving complex reasoning tasks by Reinforcement Learning with Verifiable Rewards (RLVR). This advancement is also inseparable from the oversight automated by reliable verifiers. However, current outcome-based verifiers (OVs) are unable to inspect the unreliable intermediate steps in the long reasoning chains of thought (CoTs). Meanwhile, current process-based verifiers (PVs) have difficulties in reliably detecting errors in the complex long CoTs, limited by the scarcity of high-quality annotations due to the prohibitive costs of human annotations. Therefore, we propose the \textbf{O}utcome-based \textbf{P}rocess \textbf{V}erifier (OPV), which verifies the rationale process of summarized outcomes from long CoTs to achieve both accurate and efficient verification and enable large-scale annotation. To empower the proposed verifier, we adopt an iterative active learning framework with expert annotations to progressively improve the verification capability of OPV with fewer annotation costs. Specifically, in each iteration, the most uncertain cases of the current best OPV are annotated and then subsequently used to train a new OPV through Rejection Fine-Tuning (RFT) and RLVR for the next round. Extensive experiments demonstrate OPV's superior performance and broad applicability. It achieves new state-of-the-art results on our held-out \textsc{\thisbench}, outperforming much larger open-source models such as Qwen3-Max-Preview with an F1 score of 83.1 compared to 76.3. Furthermore, OPV effectively detects false positives within synthetic dataset, closely align with expert assessment. When collaborating with policy models, OPV consistently yields performance gains, e.g., raising the accuracy of DeepSeek-R1-Distill-Qwen-32B from 55.2\% to 73.3\% on AIME2025 as the compute budget scales.
Large Language Model (LLM)-based code assistants have emerged as a powerful application of generative AI, demonstrating impressive capabilities in code generation and comprehension. A key requirement for these systems is their ability to accurately follow user instructions. We present Precise Automatically Checked Instruction Following In Code (PACIFIC), a novel framework designed to automatically generate benchmarks that rigorously assess sequential instruction-following and code dry-running capabilities in LLMs, while allowing control over benchmark difficulty. PACIFIC produces benchmark variants with clearly defined expected outputs, enabling straightforward and reliable evaluation through simple output comparisons. In contrast to existing approaches that often rely on tool usage or agentic behavior, our work isolates and evaluates the LLM's intrinsic ability to reason through code behavior step-by-step without execution (dry running) and to follow instructions. Furthermore, our framework mitigates training data contamination by facilitating effortless generation of novel benchmark variations. We validate our framework by generating a suite of benchmarks spanning a range of difficulty levels and evaluating multiple state-of-the-art LLMs. Our results demonstrate that PACIFIC can produce increasingly challenging benchmarks that effectively differentiate instruction-following and dry running capabilities, even among advanced models. Overall, our framework offers a scalable, contamination-resilient methodology for assessing core competencies of LLMs in code-related tasks.
Procedural memory enables large language model (LLM) agents to internalize "how-to" knowledge, theoretically reducing redundant trial-and-error. However, existing frameworks predominantly suffer from a "passive accumulation" paradigm, treating memory as a static append-only archive. To bridge the gap between static storage and dynamic reasoning, we propose $\textbf{ReMe}$ ($\textit{Remember Me, Refine Me}$), a comprehensive framework for experience-driven agent evolution. ReMe innovates across the memory lifecycle via three mechanisms: 1) $\textit{multi-faceted distillation}$, which extracts fine-grained experiences by recognizing success patterns, analyzing failure triggers and generating comparative insights; 2) $\textit{context-adaptive reuse}$, which tailors historical insights to new contexts via scenario-aware indexing; and 3) $\textit{utility-based refinement}$, which autonomously adds valid memories and prunes outdated ones to maintain a compact, high-quality experience pool. Extensive experiments on BFCL-V3 and AppWorld demonstrate that ReMe establishes a new state-of-the-art in agent memory system. Crucially, we observe a significant memory-scaling effect: Qwen3-8B equipped with ReMe outperforms larger, memoryless Qwen3-14B, suggesting that self-evolving memory provides a computation-efficient pathway for lifelong learning. We release our code and the $\texttt{reme.library}$ dataset to facilitate further research.
As Large Language Models (LLM) based multi-agent systems become increasingly prevalent, the collective behaviors, e.g., collective intelligence, of such artificial communities have drawn growing attention. This work aims to answer a fundamental question: How does diversity of values shape the collective behavior of AI communities? Using naturalistic value elicitation grounded in the prevalent Schwartz's Theory of Basic Human Values, we constructed multi-agent simulations where communities with varying numbers of agents engaged in open-ended interactions and constitution formation. The results show that value diversity enhances value stability, fosters emergent behaviors, and brings more creative principles developed by the agents themselves without external guidance. However, these effects also show diminishing returns: extreme heterogeneity induces instability. This work positions value diversity as a new axis of future AI capability, bridging AI ability and sociological studies of institutional emergence.
We present Thinking While Driving, a concurrent routing framework that integrates LLMs into a graph-based traffic environment. Unlike approaches that require agents to stop and deliberate, our system enables LLM-based route planning while agents are moving, significantly reducing intersection wait times. Under high traffic, agents average just 0.75 seconds of decision latency. To coordinate many agents in real-time, we implement a non-blocking asynchronous architecture using Unity coroutines and a dedicated request manager. The environment is a weighted undirected graph with live congestion metrics, updated continuously by the agents to enable shared perception. Our results show LLM-driven agents can dynamically adapt to traffic, reroute around congestion, and exhibit behaviors beyond static pathfinding, all while maintaining real-time performance. This work provides a reproducible framework for future research in adaptive routing and multi-agent cooperation.
We propose LEO-RobotAgent, a general-purpose language-driven intelligent agent framework for robots. Under this framework, LLMs can operate different types of robots to complete unpredictable complex tasks across various scenarios. This framework features strong generalization, robustness, and efficiency. The application-level system built around it can fully enhance bidirectional human-robot intent understanding and lower the threshold for human-robot interaction. Regarding robot task planning, the vast majority of existing studies focus on the application of large models in single-task scenarios and for single robot types. These algorithms often have complex structures and lack generalizability. Thus, the proposed LEO-RobotAgent framework is designed with a streamlined structure as much as possible, enabling large models to independently think, plan, and act within this clear framework. We provide a modular and easily registrable toolset, allowing large models to flexibly call various tools to meet different requirements. Meanwhile, the framework incorporates a human-robot interaction mechanism, enabling the algorithm to collaborate with humans like a partner. Experiments have verified that this framework can be easily adapted to mainstream robot platforms including unmanned aerial vehicles (UAVs), robotic arms, and wheeled robot, and efficiently execute a variety of carefully designed tasks with different complexity levels. Our code is available at https://github.com/LegendLeoChen/LEO-RobotAgent.
Large language model (LLM) agents exhibit strong mathematical problem-solving abilities and can even solve International Mathematical Olympiad (IMO) level problems with the assistance of formal proof systems. However, due to weak heuristics for auxiliary constructions, AI for geometry problem solving remains dominated by expert models such as AlphaGeometry 2, which rely heavily on large-scale data synthesis and search for both training and evaluation. In this work, we make the first attempt to build a medalist-level LLM agent for geometry and present InternGeometry. InternGeometry overcomes the heuristic limitations in geometry by iteratively proposing propositions and auxiliary constructions, verifying them with a symbolic engine, and reflecting on the engine's feedback to guide subsequent proposals. A dynamic memory mechanism enables InternGeometry to conduct more than two hundred interactions with the symbolic engine per problem. To further accelerate learning, we introduce Complexity-Boosting Reinforcement Learning (CBRL), which gradually increases the complexity of synthesized problems across training stages. Built on InternThinker-32B, InternGeometry solves 44 of 50 IMO geometry problems (2000-2024), exceeding the average gold medalist score (40.9), using only 13K training examples, just 0.004% of the data used by AlphaGeometry 2, demonstrating the potential of LLM agents on expert-level geometry tasks. InternGeometry can also propose novel auxiliary constructions for IMO problems that do not appear in human solutions. We will release the model, data, and symbolic engine to support future research.
Procedural Content Generation (PCG) offers scalable methods for algorithmically creating complex, customizable worlds. However, controlling these pipelines requires the precise configuration of opaque technical parameters. We propose a training-free architecture that utilizes LLM agents for zero-shot PCG parameter configuration. While Large Language Models (LLMs) promise a natural language interface for PCG tools, off-the-shelf models often fail to bridge the semantic gap between abstract user instructions and strict parameter specifications. Our system pairs an Actor agent with a Critic agent, enabling an iterative workflow where the system autonomously reasons over tool parameters and refines configurations to progressively align with human design preferences. We validate this approach on the generation of various 3D maps, establishing a new benchmark for instruction-following in PCG. Experiments demonstrate that our approach outperforms single-agent baselines, producing diverse and structurally valid environments from natural language descriptions. These results demonstrate that off-the-shelf LLMs can be effectively repurposed as generalized agents for arbitrary PCG tools. By shifting the burden from model training to architectural reasoning, our method offers a scalable framework for mastering complex software without task-specific fine-tuning.
This paper presents a psychologically-aware conversational agent designed to enhance both learning performance and emotional well-being in educational settings. The system combines Large Language Models (LLMs), a knowledge graph-enhanced BERT (KG-BERT), and a bidirectional Long Short-Term Memory (LSTM) with attention to classify students' cognitive and affective states in real time. Unlike prior chatbots limited to either tutoring or affective support, our approach leverages multimodal data-including textual semantics, prosodic speech features, and temporal behavioral trends-to infer engagement, stress, and conceptual understanding. A pilot study with university students demonstrated improved motivation, reduced stress, and moderate academic gains compared to baseline methods. These results underline the promise of integrating semantic reasoning, multimodal fusion, and temporal modeling to support adaptive, student-centered educational interventions.
LLM-agent based binary code analysis has demonstrated significant potential across a wide range of software security scenarios, including vulnerability detection, malware analysis, etc. In agent workflow, however, retrieving the positive from thousands of stripped binary functions based on user query remains under-studied and challenging, as the absence of symbolic information distinguishes it from source code retrieval. In this paper, we introduce, BinSeek, the first two-stage cross-modal retrieval framework for stripped binary code analysis. It consists of two models: BinSeekEmbedding is trained on large-scale dataset to learn the semantic relevance of the binary code and the natural language description, furthermore, BinSeek-Reranker learns to carefully judge the relevance of the candidate code to the description with context augmentation. To this end, we built an LLM-based data synthesis pipeline to automate training construction, also deriving a domain benchmark for future research. Our evaluation results show that BinSeek achieved the state-of-the-art performance, surpassing the the same scale models by 31.42% in Rec@3 and 27.17% in MRR@3, as well as leading the advanced general-purpose models that have 16 times larger parameters.
Agentic systems built on large language models operate through recursive feedback loops, where each output becomes the next input. Yet the geometric behavior of these agentic loops (whether they converge, diverge, or exhibit more complex dynamics) remains poorly understood. This paper introduces a geometric framework for analyzing agentic trajectories in semantic embedding space, treating iterative transformations as discrete dynamical systems. We distinguish the artifact space, where linguistic transformations occur, from the embedding space, where geometric measurements are performed. Because cosine similarity is biased by embedding anisotropy, we introduce an isotonic calibration that eliminates systematic bias and aligns similarities with human semantic judgments while preserving high local stability. This enables rigorous measurement of trajectories, clusters and attractors. Through controlled experiments on singular agentic loops, we identify two fundamental regimes. A contractive rewriting loop converges toward a stable attractor with decreasing dispersion, while an exploratory summarize and negate loop produces unbounded divergence with no cluster formation. These regimes display qualitatively distinct geometric signatures of contraction and expansion. Our results show that prompt design directly governs the dynamical regime of an agentic loop, enabling systematic control of convergence, divergence and trajectory structure in iterative LLM transformations.
Epidemic response planning is essential yet traditionally reliant on labor-intensive manual methods. This study aimed to design and evaluate EpiPlanAgent, an agent-based system using large language models (LLMs) to automate the generation and validation of digital emergency response plans. The multi-agent framework integrated task decomposition, knowledge grounding, and simulation modules. Public health professionals tested the system using real-world outbreak scenarios in a controlled evaluation. Results demonstrated that EpiPlanAgent significantly improved the completeness and guideline alignment of plans while drastically reducing development time compared to manual workflows. Expert evaluation confirmed high consistency between AI-generated and human-authored content. User feedback indicated strong perceived utility. In conclusion, EpiPlanAgent provides an effective, scalable solution for intelligent epidemic response planning, demonstrating the potential of agentic AI to transform public health preparedness.
SWE-Bench-Verified, a dataset comprising 500 issues, serves as a de facto benchmark for evaluating various large language models (LLMs) on their ability to resolve GitHub issues. But this benchmark may overlap with model training data. If that is true, scores may reflect training recall, not issue-solving skill. To study this, we test two Claude models that frequently appear in top-performing agents submitted to the benchmark. We ask them to find relevant files using only issue text, and then issue text plus file paths. We then run the same setup on BeetleBox and SWE-rebench. Despite both benchmarks involving popular open-source Python projects, models performed 3 times better on SWE-Bench-Verified. They were also 6 times better at finding edited files, without any additional context about the projects themselves. This gap suggests the models may have seen many SWE-Bench-Verified tasks during training. As a result, scores on this benchmark may not reflect an agent's ability to handle real software issues, yet it continues to be used in ways that can misrepresent progress and lead to choices that favour agents that use certain models over strong agent design. Our setup tests the localization step with minimal context to the extent that the task should be logically impossible to solve. Our results show the risk of relying on older popular benchmarks and support the shift toward newer datasets built with contamination in mind.
Medical care follows complex clinical pathways that extend beyond isolated physician-patient encounters, emphasizing decision-making and transitions between different stages. Current benchmarks focusing on static exams or isolated dialogues inadequately evaluate large language models (LLMs) in dynamic clinical scenarios. We introduce CP-Env, a controllable agentic hospital environment designed to evaluate LLMs across end-to-end clinical pathways. CP-Env simulates a hospital ecosystem with patient and physician agents, constructing scenarios ranging from triage and specialist consultation to diagnostic testing and multidisciplinary team meetings for agent interaction. Following real hospital adaptive flow of healthcare, it enables branching, long-horizon task execution. We propose a three-tiered evaluation framework encompassing Clinical Efficacy, Process Competency, and Professional Ethics. Results reveal that most models struggle with pathway complexity, exhibiting hallucinations and losing critical diagnostic details. Interestingly, excessive reasoning steps can sometimes prove counterproductive, while top models tend to exhibit reduced tool dependency through internalized knowledge. CP-Env advances medical AI agents development through comprehensive end-to-end clinical evaluation. We provide the benchmark and evaluation tools for further research and development at https://github.com/SPIRAL-MED/CP-Env.
Evaluating large language models (LLMs) has recently emerged as a critical issue for safe and trustworthy application of LLMs in the medical domain. Although a variety of static medical question-answering (QA) benchmarks have been proposed, many aspects remain underexplored, such as the effectiveness of LLMs in generating responses in dynamic, interactive clinical multi-turn conversation situations and the identification of multi-faceted evaluation strategies beyond simple accuracy. However, formally evaluating a dynamic, interactive clinical situation is hindered by its vast combinatorial space of possible patient states and interaction trajectories, making it difficult to standardize and quantitatively measure such scenarios. Here, we introduce AutoMedic, a multi-agent simulation framework that enables automated evaluation of LLMs as clinical conversational agents. AutoMedic transforms off-the-shelf static QA datasets into virtual patient profiles, enabling realistic and clinically grounded multi-turn clinical dialogues between LLM agents. The performance of various clinical conversational agents is then assessed based on our CARE metric, which provides a multi-faceted evaluation standard of clinical conversational accuracy, efficiency/strategy, empathy, and robustness. Our findings, validated by human experts, demonstrate the validity of AutoMedic as an automated evaluation framework for clinical conversational agents, offering practical guidelines for the effective development of LLMs in conversational medical applications.
Central to long-form text generation in vertical domains is the "impossible trinity" confronting current large language models (LLMs): the simultaneous achievement of low hallucination, deep logical coherence, and personalized expression. This study establishes that this bottleneck arises from existing generative paradigms succumbing to the Statistical Smoothing Trap, a phenomenon that overlooks the high-entropy information acquisition and structured cognitive processes integral to expert-level writing. To address this limitation, we propose the DeepNews Framework, an agentic workflow that explicitly models the implicit cognitive processes of seasoned financial journalists. The framework integrates three core modules: first, a dual-granularity retrieval mechanism grounded in information foraging theory, which enforces a 10:1 saturated information input ratio to mitigate hallucinatory outputs; second, schema-guided strategic planning, a process leveraging domain expert knowledge bases (narrative schemas) and Atomic Blocks to forge a robust logical skeleton; third, adversarial constraint prompting, a technique deploying tactics including Rhythm Break and Logic Fog to disrupt the probabilistic smoothness inherent in model-generated text. Experiments delineate a salient Knowledge Cliff in deep financial reporting: content truthfulness collapses when retrieved context falls below 15,000 characters, while a high-redundancy input exceeding 30,000 characters stabilizes the Hallucination-Free Rate (HFR) above 85%. In an ecological validity blind test conducted with a top-tier Chinese technology media outlet, the DeepNews system--built on a previous-generation model (DeepSeek-V3-0324)-achieved a 25% submission acceptance rate, significantly outperforming the 0% acceptance rate of zero-shot generation by a state-of-the-art (SOTA) model (GPT-5).
Large language model (LLM)-driven agents are emerging as a powerful new paradigm for solving complex problems. Despite the empirical success of these practices, a theoretical framework to understand and unify their macroscopic dynamics remains lacking. This Letter proposes a method based on the least action principle to estimate the underlying generative directionality of LLMs embedded within agents. By experimentally measuring the transition probabilities between LLM-generated states, we statistically discover a detailed balance in LLM-generated transitions, indicating that LLM generation may not be achieved by generally learning rule sets and strategies, but rather by implicitly learning a class of underlying potential functions that may transcend different LLM architectures and prompt templates. To our knowledge, this is the first discovery of a macroscopic physical law in LLM generative dynamics that does not depend on specific model details. This work is an attempt to establish a macroscopic dynamics theory of complex AI systems, aiming to elevate the study of AI agents from a collection of engineering practices to a science built on effective measurements that are predictable and quantifiable.
Force field-based molecular dynamics (MD) simulations are indispensable for probing the structure, dynamics, and functions of biomolecular systems, including proteins and protein-ligand complexes. Despite their broad utility in drug discovery and protein engineering, the technical complexity of MD setup, encompassing parameterization, input preparation, and software configuration, remains a major barrier for widespread and efficient usage. Agentic LLMs have demonstrated their capacity to autonomously execute multi-step scientific processes, and to date, they have not successfully been used to automate protein-ligand MD workflows. Here, we present DynaMate, a modular multi-agent framework that autonomously designs and executes complete MD workflows for both protein and protein-ligand systems, and offers free energy binding affinity calculations with the MM/PB(GB)SA method. The framework integrates dynamic tool use, web search, PaperQA, and a self-correcting behavior. DynaMate comprises three specialized modules, interacting to plan the experiment, perform the simulation, and analyze the results. We evaluated its performance across twelve benchmark systems of varying complexity, assessing success rate, efficiency, and adaptability. DynaMate reliably performed full MD simulations, corrected runtime errors through iterative reasoning, and produced meaningful analyses of protein-ligand interactions. This automated framework paves the way toward standardized, scalable, and time-efficient molecular modeling pipelines for future biomolecular and drug design applications.
Long-term planning in complex, text-based environments presents significant challenges due to open-ended action spaces, ambiguous observations, and sparse feedback. Recent research suggests that large language models (LLMs) encode rich semantic knowledge about the world, which can be valuable for guiding agents in high-level reasoning and planning across both embodied and purely textual settings. However, existing approaches often depend heavily on querying LLMs during training and inference, making them computationally expensive and difficult to deploy efficiently. In addition, these methods typically employ a pretrained, unaltered LLM whose parameters remain fixed throughout training, providing no opportunity for adaptation to the target task. To address these limitations, we introduce SCOPE (Subgoal-COnditioned Pretraining for Efficient planning), a one-shot hierarchical planner that leverages LLM-generated subgoals only at initialization to pretrain a lightweight student model. Unlike prior approaches that distill LLM knowledge by repeatedly prompting the model to adaptively generate subgoals during training, our method derives subgoals directly from example trajectories. This design removes the need for repeated LLM queries, significantly improving efficiency, though at the cost of reduced explainability and potentially suboptimal subgoals. Despite their suboptimality, our results on the TextCraft environment show that LLM-generated subgoals can still serve as a strong starting point for hierarchical goal decomposition in text-based planning tasks. Compared to the LLM-based hierarchical agent ADaPT (Prasad et al., 2024), which achieves a 0.52 success rate, our method reaches 0.56 and reduces inference time from 164.4 seconds to just 3.0 seconds.
Culture is a core component of human-to-human interaction and plays a vital role in how we perceive and interact with others. Advancements in the effectiveness of Large Language Models (LLMs) in generating human-sounding text have greatly increased the amount of human-to-computer interaction. As this field grows, the cultural alignment of these human-like agents becomes an important field of study. Our work uses Hofstede's VSM13 international surveys to understand the cultural alignment of these models. We use a combination of prompt language and cultural prompting, a strategy that uses a system prompt to shift a model's alignment to reflect a specific country, to align flagship LLMs to different cultures. Our results show that DeepSeek-V3, V3.1, and OpenAI's GPT-5 exhibit a close alignment with the survey responses of the United States and do not achieve a strong or soft alignment with China, even when using cultural prompts or changing the prompt language. We also find that GPT-4 exhibits an alignment closer to China when prompted in English, but cultural prompting is effective in shifting this alignment closer to the United States. Other low-cost models, GPT-4o and GPT-4.1, respond to the prompt language used (i.e., English or Simplified Chinese) and cultural prompting strategies to create acceptable alignments with both the United States and China.
We present an end-to-end framework for planning supported by verifiers. An orchestrator receives a human specification written in natural language and converts it into a PDDL (Planning Domain Definition Language) model, where the domain and problem are iteratively refined by sub-modules (agents) to address common planning requirements, such as time constraints and optimality, as well as ambiguities and contradictions that may exist in the human specification. The validated domain and problem are then passed to an external planning engine to generate a plan. The orchestrator and agents are powered by Large Language Models (LLMs) and require no human intervention at any stage of the process. Finally, a module translates the final plan back into natural language to improve human readability while maintaining the correctness of each step. We demonstrate the flexibility and effectiveness of our framework across various domains and tasks, including the Google NaturalPlan benchmark and PlanBench, as well as planning problems like Blocksworld and the Tower of Hanoi (where LLMs are known to struggle even with small instances). Our framework can be integrated with any PDDL planning engine and validator (such as Fast Downward, LPG, POPF, VAL, and uVAL, which we have tested) and represents a significant step toward end-to-end planning aided by LLMs.
We present Auto-BenchmarkCard, a workflow for generating validated descriptions of AI benchmarks. Benchmark documentation is often incomplete or inconsistent, making it difficult to interpret and compare benchmarks across tasks or domains. Auto-BenchmarkCard addresses this gap by combining multi-agent data extraction from heterogeneous sources (e.g., Hugging Face, Unitxt, academic papers) with LLM-driven synthesis. A validation phase evaluates factual accuracy through atomic entailment scoring using the FactReasoner tool. This workflow has the potential to promote transparency, comparability, and reusability in AI benchmark reporting, enabling researchers and practitioners to better navigate and evaluate benchmark choices.
The rise of multi-agent systems powered by large language models (LLMs) and specialized reasoning agents exposes fundamental limitations in today's data management architectures. Traditional databases and data fabrics were designed for static, well-defined workloads, whereas agentic systems exhibit dynamic, context-driven, and collaborative behaviors. Agents continuously decompose tasks, shift attention across modalities, and share intermediate results with peers - producing non-deterministic, multi-modal workloads that strain conventional query optimizers and caching mechanisms. We propose an Agent-Centric Data Fabric, a unified architecture that rethinks how data systems serve, optimize, coordinate, and learn from agentic workloads. To achieve this we exploit the concepts of attention-guided data retrieval, semantic micro-caching for context-driven agent federations, predictive data prefetching and quorum-based data serving. Together, these mechanisms enable agents to access representative data faster and more efficiently, while reducing redundant queries, data movement, and inference load across systems. By framing data systems as adaptive collaborators, instead of static executors, we outline new research directions toward behaviorally responsive data infrastructures, where caching, probing, and orchestration jointly enable efficient, context-rich data exchange among dynamic, reasoning-driven agents.
Context. LLM-based autonomous agents in software engineering rely on large, proprietary models, limiting local deployment. This has spurred interest in Small Language Models (SLMs), but their practical effectiveness and efficiency within complex agentic frameworks for automated issue resolution remain poorly understood. Goal. We investigate the performance, energy efficiency, and resource consumption of four leading agentic issue resolution frameworks when deliberately constrained to using SLMs. We aim to assess the viability of these systems for this task in resource-limited settings and characterize the resulting trade-offs. Method. We conduct a controlled evaluation of four leading agentic frameworks (SWE-Agent, OpenHands, Mini SWE Agent, AutoCodeRover) using two SLMs (Gemma-3 4B, Qwen-3 1.7B) on the SWE-bench Verified Mini benchmark. On fixed hardware, we measure energy, duration, token usage, and memory over 150 runs per configuration. Results. We find that framework architecture is the primary driver of energy consumption. The most energy-intensive framework, AutoCodeRover (Gemma), consumed 9.4x more energy on average than the least energy-intensive, OpenHands (Gemma). However, this energy is largely wasted. Task resolution rates were near-zero, demonstrating that current frameworks, when paired with SLMs, consume significant energy on unproductive reasoning loops. The SLM's limited reasoning was the bottleneck for success, but the framework's design was the bottleneck for efficiency. Conclusions. Current agentic frameworks, designed for powerful LLMs, fail to operate efficiently with SLMs. We find that framework architecture is the primary driver of energy consumption, but this energy is largely wasted due to the SLMs' limited reasoning. Viable low-energy solutions require shifting from passive orchestration to architectures that actively manage SLM weaknesses.
Prompt injection attacks aim to contaminate the input data of an LLM to mislead it into completing an attacker-chosen task instead of the intended task. In many applications and agents, the input data originates from multiple sources, with each source contributing a segment of the overall input. In these multi-source scenarios, an attacker may control only a subset of the sources and contaminate the corresponding segments, but typically does not know the order in which the segments are arranged within the input. Existing prompt injection attacks either assume that the entire input data comes from a single source under the attacker's control or ignore the uncertainty in the ordering of segments from different sources. As a result, their success is limited in domains involving multi-source data. In this work, we propose ObliInjection, the first prompt injection attack targeting LLM applications and agents with multi-source input data. ObliInjection introduces two key technical innovations: the order-oblivious loss, which quantifies the likelihood that the LLM will complete the attacker-chosen task regardless of how the clean and contaminated segments are ordered; and the orderGCG algorithm, which is tailored to minimize the order-oblivious loss and optimize the contaminated segments. Comprehensive experiments across three datasets spanning diverse application domains and twelve LLMs demonstrate that ObliInjection is highly effective, even when only one out of 6-100 segments in the input data is contaminated.
Next-Generation Sequencing (NGS) has become a cornerstone of genomic research, yet the complexity of downstream analysis-ranging from differential expression gene (DEG) identification to biological interpretations-remains a significant barrier for researchers lacking specialized computational and biological expertise. While recent studies have introduced AI agents for RNA-seq analysis, most focus on general workflows without offering tailored interpretations or guidance for novices. To address this gap, we developed an Agentic AI model designed to automate NGS downstream analysis, provide literature-backed interpretations, and autonomously recommend advanced analytical methods. Built on the Llama 3 70B Large Language Model (LLM) and a Retrieval-Augmented Generation (RAG) framework, the model is deployed as an interactive Streamlit web application. The system integrates standard bioinformatics tools (Biopython, GSEApy, gProfiler) to execute core analyses, including DEG identification, clustering, and pathway enrichment. Uniquely, the agent utilizes RAG to query PubMed via Entrez, synthesizing biological insights and validating hypotheses with current literature. In a case study using cancer-related dataset, the model successfully identified significant DEGs, visualized clinical correlations, and derived evidence-based insights (e.g., linking BRAF mutations to prognosis), subsequently executing advanced survival modeling upon user selection. This framework democratizes bioinformatics by enabling researchers with limited backgrounds to seamlessly transition from basic data processing to advanced hypothesis testing and validation.
Large language models (LLMs) are increasingly being deployed as autonomous agents on behalf of institutions and individuals in economic, political, and social settings that involve negotiation. Yet this trend carries significant risks if their strategic behavior is not well understood. In this work, we revisit the NegotiationArena framework and run controlled simulation experiments on a diverse set of frontier LLMs across three multi turn bargaining games: Buyer Seller, Multi turn Ultimatum, and Resource Exchange. We ask whether improved general reasoning capabilities lead to rational, unbiased, and convergent negotiation strategies. Our results challenge this assumption. We find that models diverge into distinct, model specific strategic equilibria rather than converging to a unified optimal behavior. Moreover, strong numerical and semantic anchoring effects persist: initial offers are highly predictive of final agreements, and models consistently generate biased proposals by collapsing diverse internal valuations into rigid, generic price points. More concerningly, we observe dominance patterns in which some models systematically achieve higher payoffs than their counterparts. These findings underscore an urgent need to develop mechanisms to mitigate these issues before deploying such systems in real-world scenarios.
Deception is a fundamental challenge for multi-agent reasoning: effective systems must strategically conceal information while detecting misleading behavior in others. Yet most evaluations reduce deception to static classification, ignoring the interactive, adversarial, and longitudinal nature of real deceptive dynamics. Large language models (LLMs) can deceive convincingly but remain weak at detecting deception in peers. We present WOLF, a multi-agent social deduction benchmark based on Werewolf that enables separable measurement of deception production and detection. WOLF embeds role-grounded agents (Villager, Werewolf, Seer, Doctor) in a programmable LangGraph state machine with strict night-day cycles, debate turns, and majority voting. Every statement is a distinct analysis unit, with self-assessed honesty from speakers and peer-rated deceptiveness from others. Deception is categorized via a standardized taxonomy (omission, distortion, fabrication, misdirection), while suspicion scores are longitudinally smoothed to capture both immediate judgments and evolving trust dynamics. Structured logs preserve prompts, outputs, and state transitions for full reproducibility. Across 7,320 statements and 100 runs, Werewolves produce deceptive statements in 31% of turns, while peer detection achieves 71-73% precision with ~52% overall accuracy. Precision is higher for identifying Werewolves, though false positives occur against Villagers. Suspicion toward Werewolves rises from ~52% to over 60% across rounds, while suspicion toward Villagers and the Doctor stabilizes near 44-46%. This divergence shows that extended interaction improves recall against liars without compounding errors against truthful roles. WOLF moves deception evaluation beyond static datasets, offering a dynamic, controlled testbed for measuring deceptive and detective capacity in adversarial multi-agent interaction.