Large Language Models (LLMs) have emerged as powerful operators for evolutionary search, yet the design of efficient search scaffolds remains ad hoc. While promising, current LLM-in-the-loop systems lack a systematic approach to managing the evolutionary process. We identify three distinct failure modes: Context Pollution, where experiment history biases future candidate generation; Mode Collapse, where agents stagnate in local minima due to poor exploration-exploitation balance; and Weak Collaboration, where rigid crossover strategies fail to leverage parallel search trajectories effectively. We introduce Progress-Aware Consistent Evolution (PACEvolve), a framework designed to robustly govern the agent's context and search dynamics, to address these challenges. PACEvolve combines hierarchical context management (HCM) with pruning to address context pollution; momentum-based backtracking (MBB) to escape local minima; and a self-adaptive sampling policy that unifies backtracking and crossover for dynamic search coordination (CE), allowing agents to balance internal refinement with cross-trajectory collaboration. We demonstrate that PACEvolve provides a systematic path to consistent, long-horizon self-improvement, achieving state-of-the-art results on LLM-SR and KernelBench, while discovering solutions surpassing the record on Modded NanoGPT.
As LLM-based systems increasingly operate as agents embedded within human social and technical systems, alignment can no longer be treated as a property of an isolated model, but must be understood in relation to the environments in which these agents act. Even the most sophisticated methods of alignment, such as Reinforcement Learning through Human Feedback (RHLF) or through AI Feedback (RLAIF) cannot ensure control once internal goal structures diverge from developer intent. We identify three structural problems that emerge from core properties of AI models: (1) behavioral goal-independence, where models develop internal objectives and misgeneralize goals; (2) instrumental override of natural-language constraints, where models regard safety principles as non-binding while pursuing latent objectives, leveraging deception and manipulation; and (3) agentic alignment drift, where individually aligned agents converge to collusive equilibria through interaction dynamics invisible to single-agent audits. The solution this paper advances is Institutional AI: a system-level approach that treats alignment as a question of effective governance of AI agent collectives. We argue for a governance-graph that details how to constrain agents via runtime monitoring, incentive shaping through prizes and sanctions, explicit norms and enforcement roles. This institutional turn reframes safety from software engineering to a mechanism design problem, where the primary goal of alignment is shifting the payoff landscape of AI agent collectives.
Comprehending genomic information is essential for biomedical research, yet extracting data from complex distributed databases remains challenging. Large language models (LLMs) offer potential for genomic Question Answering (QA) but face limitations due to restricted access to domain-specific databases. GeneGPT is the current state-of-the-art system that enhances LLMs by utilizing specialized API calls, though it is constrained by rigid API dependencies and limited adaptability. We replicate GeneGPT and propose GenomAgent, a multi-agent framework that efficiently coordinates specialized agents for complex genomics queries. Evaluated on nine tasks from the GeneTuring benchmark, GenomAgent outperforms GeneGPT by 12% on average, and its flexible architecture extends beyond genomics to various scientific domains needing expert knowledge extraction.
In this article, we argue that understanding the collective behavior of agents based on large language models (LLMs) is an essential area of inquiry, with important implications in terms of risks and benefits, impacting us as a society at many levels. We claim that the distinctive nature of LLMs--namely, their initialization with extensive pre-trained knowledge and implicit social priors, together with their capability of adaptation through in-context learning--motivates the need for an interactionist paradigm consisting of alternative theoretical foundations, methodologies, and analytical tools, in order to systematically examine how prior knowledge and embedded values interact with social context to shape emergent phenomena in multi-agent generative AI systems. We propose and discuss four directions that we consider crucial for the development and deployment of LLM-based collectives, focusing on theory, methods, and trans-disciplinary dialogue.
As AI agents become increasingly autonomous, widely deployed in consequential contexts, and efficacious in bringing about real-world impacts, ensuring that their decisions are not only instrumentally effective but also normatively aligned has become critical. We introduce a neuro-symbolic reason-based containment architecture, Governor for Reason-Aligned ContainmEnt (GRACE), that decouples normative reasoning from instrumental decision-making and can contain AI agents of virtually any design. GRACE restructures decision-making into three modules: a Moral Module (MM) that determines permissible macro actions via deontic logic-based reasoning; a Decision-Making Module (DMM) that encapsulates the target agent while selecting instrumentally optimal primitive actions in accordance with derived macro actions; and a Guard that monitors and enforces moral compliance. The MM uses a reason-based formalism providing a semantic foundation for deontic logic, enabling interpretability, contestability, and justifiability. Its symbolic representation enriches the DMM's informational context and supports formal verification and statistical guarantees of alignment enforced by the Guard. We demonstrate GRACE on an example of a LLM therapy assistant, showing how it enables stakeholders to understand, contest, and refine agent behavior.
As Large Language Models (LLMs) increasingly operate as Deep Research (DR) Agents capable of autonomous investigation and information synthesis, reliable evaluation of their task performance has become a critical bottleneck. Current benchmarks predominantly rely on static datasets, which suffer from several limitations: limited task generality, temporal misalignment, and data contamination. To address these, we introduce DR-Arena, a fully automated evaluation framework that pushes DR agents to their capability limits through dynamic investigation. DR-Arena constructs real-time Information Trees from fresh web trends to ensure the evaluation rubric is synchronized with the live world state, and employs an automated Examiner to generate structured tasks testing two orthogonal capabilities: Deep reasoning and Wide coverage. DR-Arena further adopts Adaptive Evolvement Loop, a state-machine controller that dynamically escalates task complexity based on real-time performance, demanding deeper deduction or wider aggregation until a decisive capability boundary emerges. Experiments with six advanced DR agents demonstrate that DR-Arena achieves a Spearman correlation of 0.94 with the LMSYS Search Arena leaderboard. This represents the state-of-the-art alignment with human preferences without any manual efforts, validating DR-Arena as a reliable alternative for costly human adjudication.
The advancement of artificial intelligence toward agentic science is currently bottlenecked by the challenge of ultra-long-horizon autonomy, the ability to sustain strategic coherence and iterative correction over experimental cycles spanning days or weeks. While Large Language Models (LLMs) have demonstrated prowess in short-horizon reasoning, they are easily overwhelmed by execution details in the high-dimensional, delayed-feedback environments of real-world research, failing to consolidate sparse feedback into coherent long-term guidance. Here, we present ML-Master 2.0, an autonomous agent that masters ultra-long-horizon machine learning engineering (MLE) which is a representative microcosm of scientific discovery. By reframing context management as a process of cognitive accumulation, our approach introduces Hierarchical Cognitive Caching (HCC), a multi-tiered architecture inspired by computer systems that enables the structural differentiation of experience over time. By dynamically distilling transient execution traces into stable knowledge and cross-task wisdom, HCC allows agents to decouple immediate execution from long-term experimental strategy, effectively overcoming the scaling limits of static context windows. In evaluations on OpenAI's MLE-Bench under 24-hour budgets, ML-Master 2.0 achieves a state-of-the-art medal rate of 56.44%. Our findings demonstrate that ultra-long-horizon autonomy provides a scalable blueprint for AI capable of autonomous exploration beyond human-precedent complexities.
Advanced manufacturing with new bio-derived materials can be achieved faster and more economically with first-principle-based artificial intelligence and machine learning (AI/ML)-derived models and process optimization. Not only is this motivated by increased industry profitability, but it can also be optimized to reduce waste generation, energy consumption, and gas emissions through additive manufacturing (AM) and AI/ML-directed self-driving laboratory (SDL) process optimization. From this perspective, the benefits of using 3D printing technology to manufacture durable, sustainable materials will enable high-value reuse and promote a better circular economy. Using AI/ML workflows at different levels, it is possible to optimize the synthesis and adaptation of new bio-derived materials with self-correcting 3D printing methods, and in-situ characterization. Working with training data and hypotheses derived from Large Language Models (LLMs) and algorithms, including ML-optimized simulation, it is possible to demonstrate more field convergence. The combination of SDL and AI/ML Workflows can be the norm for improved use of biobased and renewable materials towards advanced manufacturing. This should result in faster and better structure, composition, processing, and properties (SCPP) correlation. More agentic AI tasks, as well as supervised or unsupervised learning, can be incorporated to improve optimization protocols continuously. Deep Learning (DL), Reinforcement Learning (RL), and Deep Reinforcement Learning (DRL) with Deep Neural Networks (DNNs) can be applied to more generative AI directions in both AM and SDL, with bio-based materials.
Enabling Large Language Models (LLMs) to effectively utilize tools in multi-turn interactions is essential for building capable autonomous agents. However, acquiring diverse and realistic multi-turn tool-use data remains a significant challenge. In this work, we propose a novel text-based paradigm. We observe that textual corpora naturally contain rich, multi-step problem-solving experiences, which can serve as an untapped, scalable, and authentic data source for multi-turn tool-use tasks. Based on this insight, we introduce GEM, a data synthesis pipeline that enables the generation and extraction of multi-turn tool-use trajectories from text corpora through a four-stage process: relevance filtering, workflow & tool extraction, trajectory grounding, and complexity refinement. To reduce the computational cost, we further train a specialized Trajectory Synthesizer via supervised fine-tuning. This model distills the complex generation pipeline into an efficient, end-to-end trajectory generator. Experiments demonstrate that our GEM-32B achieve a 16.5% improvement on the BFCL V3 Multi-turn benchmark. Our models partially surpass the performance of models trained on τ - bench (Airline and Retail) in-domain data, highlighting the superior generalization capability derived from our text-based synthesis paradigm. Notably, our Trajectory Synthesizer matches the quality of the full pipeline while significantly reducing inference latency and costs.
Modern coding scaffolds turn LLMs into capable software agents, but their ability to follow scaffold-specified instructions remains under-examined, especially when constraints are heterogeneous and persist across interactions. To fill this gap, we introduce OctoBench, which benchmarks scaffold-aware instruction following in repository-grounded agentic coding. OctoBench includes 34 environments and 217 tasks instantiated under three scaffold types, and is paired with 7,098 objective checklist items. To disentangle solving the task from following the rules, we provide an automated observation-and-scoring toolkit that captures full trajectories and performs fine-grained checks. Experiments on eight representative models reveal a systematic gap between task-solving and scaffold-aware compliance, underscoring the need for training and evaluation that explicitly targets heterogeneous instruction following. We release the benchmark to support reproducible benchmarking and to accelerate the development of more scaffold-aware coding agents.
The rise of AI agent frameworks has introduced agent skills, modular packages containing instructions and executable code that dynamically extend agent capabilities. While this architecture enables powerful customization, skills execute with implicit trust and minimal vetting, creating a significant yet uncharacterized attack surface. We conduct the first large-scale empirical security analysis of this emerging ecosystem, collecting 42,447 skills from two major marketplaces and systematically analyzing 31,132 using SkillScan, a multi-stage detection framework integrating static analysis with LLM-based semantic classification. Our findings reveal pervasive security risks: 26.1% of skills contain at least one vulnerability, spanning 14 distinct patterns across four categories: prompt injection, data exfiltration, privilege escalation, and supply chain risks. Data exfiltration (13.3%) and privilege escalation (11.8%) are most prevalent, while 5.2% of skills exhibit high-severity patterns strongly suggesting malicious intent. We find that skills bundling executable scripts are 2.12x more likely to contain vulnerabilities than instruction-only skills (OR=2.12, p<0.001). Our contributions include: (1) a grounded vulnerability taxonomy derived from 8,126 vulnerable skills, (2) a validated detection methodology achieving 86.7% precision and 82.5% recall, and (3) an open dataset and detection toolkit to support future research. These results demonstrate an urgent need for capability-based permission systems and mandatory security vetting before this attack vector is further exploited.
Large Language Models (LLMs) have demonstrated remarkable capabilities in reasoning and generation, serving as the foundation for advanced persona simulation and Role-Playing Language Agents (RPLAs). However, achieving authentic alignment with human cognitive and behavioral patterns remains a critical challenge for these agents. We present HUMANLLM, a framework treating psychological patterns as interacting causal forces. We construct 244 patterns from ~12,000 academic papers and synthesize 11,359 scenarios where 2-5 patterns reinforce, conflict, or modulate each other, with multi-turn conversations expressing inner thoughts, actions, and dialogue. Our dual-level checklists evaluate both individual pattern fidelity and emergent multi-pattern dynamics, achieving strong human alignment (r=0.91) while revealing that holistic metrics conflate simulation accuracy with social desirability. HUMANLLM-8B outperforms Qwen3-32B on multi-pattern dynamics despite 4x fewer parameters, demonstrating that authentic anthropomorphism requires cognitive modeling--simulating not just what humans do, but the psychological processes generating those behaviors.
We demonstrate that large language model (LLM) agents can autonomously perform tensor network simulations of quantum many-body systems, achieving approximately 90% success rate across representative benchmark tasks. Tensor network methods are powerful tools for quantum simulation, but their effective use requires expertise typically acquired through years of graduate training. By combining in-context learning with curated documentation and multi-agent decomposition, we create autonomous AI agents that can be trained in specialized computational domains within minutes. We benchmark three configurations (baseline, single-agent with in-context learning, and multi-agent with in-context learning) on problems spanning quantum phase transitions, open quantum system dynamics, and photochemical reactions. Systematic evaluation using DeepSeek-V3.2, Gemini 2.5 Pro, and Claude Opus 4.5 demonstrates that both in-context learning and multi-agent architecture are essential. Analysis of failure modes reveals characteristic patterns across models, with the multi-agent configuration substantially reducing implementation errors and hallucinations compared to simpler architectures.
Large Language Models (LLMs) have enabled the development of powerful agentic systems capable of automating complex workflows across various fields. However, these systems are highly vulnerable to indirect prompt injection attacks, where malicious instructions embedded in external data can hijack agent behavior. In this work, we present ReasAlign, a model-level solution to improve safety alignment against indirect prompt injection attacks. The core idea of ReasAlign is to incorporate structured reasoning steps to analyze user queries, detect conflicting instructions, and preserve the continuity of the user's intended tasks to defend against indirect injection attacks. To further ensure reasoning logic and accuracy, we introduce a test-time scaling mechanism with a preference-optimized judge model that scores reasoning steps and selects the best trajectory. Comprehensive evaluations across various benchmarks show that ReasAlign maintains utility comparable to an undefended model while consistently outperforming Meta SecAlign, the strongest prior guardrail. On the representative open-ended CyberSecEval2 benchmark, which includes multiple prompt-injected tasks, ReasAlign achieves 94.6% utility and only 3.6% ASR, far surpassing the state-of-the-art defensive model of Meta SecAlign (56.4% utility and 74.4% ASR). These results demonstrate that ReasAlign achieves the best trade-off between security and utility, establishing a robust and practical defense against prompt injection attacks in real-world agentic systems. Our code and experimental results could be found at https://github.com/leolee99/ReasAlign.
We introduce AWED-FiNER, an open-source ecosystem designed to bridge the gap in Fine-grained Named Entity Recognition (FgNER) for 36 global languages spoken by more than 6.6 billion people. While Large Language Models (LLMs) dominate general Natural Language Processing (NLP) tasks, they often struggle with low-resource languages and fine-grained NLP tasks. AWED-FiNER provides a collection of agentic toolkits, web applications, and several state-of-the-art expert models that provides FgNER solutions across 36 languages. The agentic tools enable to route multilingual text to specialized expert models and fetch FgNER annotations within seconds. The web-based platforms provide ready-to-use FgNER annotation service for non-technical users. Moreover, the collection of language specific extremely small sized open-source state-of-the-art expert models facilitate offline deployment in resource contraint scenerios including edge devices. AWED-FiNER covers languages spoken by over 6.6 billion people, including a specific focus on vulnerable languages such as Bodo, Manipuri, Bishnupriya, and Mizo. The resources can be accessed here: Agentic Tool (https://github.com/PrachuryyaKaushik/AWED-FiNER), Web Application (https://hf.co/spaces/prachuryyaIITG/AWED-FiNER), and 49 Expert Detector Models (https://hf.co/collections/prachuryyaIITG/awed-finer).
While LLM-based agents can interact with environments via invoking external tools, their expanded capabilities also amplify security risks. Monitoring step-level tool invocation behaviors in real time and proactively intervening before unsafe execution is critical for agent deployment, yet remains under-explored. In this work, we first construct TS-Bench, a novel benchmark for step-level tool invocation safety detection in LLM agents. We then develop a guardrail model, TS-Guard, using multi-task reinforcement learning. The model proactively detects unsafe tool invocation actions before execution by reasoning over the interaction history. It assesses request harmfulness and action-attack correlations, producing interpretable and generalizable safety judgments and feedback. Furthermore, we introduce TS-Flow, a guardrail-feedback-driven reasoning framework for LLM agents, which reduces harmful tool invocations of ReAct-style agents by 65 percent on average and improves benign task completion by approximately 10 percent under prompt injection attacks.
Causal discovery aims to recover ``what causes what'', but classical constraint-based methods (e.g., PC, FCI) suffer from error propagation, and recent LLM-based causal oracles often behave as opaque, confidence-free black boxes. This paper introduces Tree-Query, a tree-structured, multi-expert LLM framework that reduces pairwise causal discovery to a short sequence of queries about backdoor paths, (in)dependence, latent confounding, and causal direction, yielding interpretable judgments with robustness-aware confidence scores. Theoretical guarantees are provided for asymptotic identifiability of four pairwise relations. On data-free benchmarks derived from Mooij et al. and UCI causal graphs, Tree-Query improves structural metrics over direct LLM baselines, and a diet--weight case study illustrates confounder screening and stable, high-confidence causal conclusions. Tree-Query thus offers a principled way to obtain data-free causal priors from LLMs that can complement downstream data-driven causal discovery. Code is available at https://anonymous.4open.science/r/Repo-9B3E-4F96.
Generating molecules that satisfy precise numeric constraints over multiple physicochemical properties is critical and challenging. Although large language models (LLMs) are expressive, they struggle with precise multi-objective control and numeric reasoning without external structure and feedback. We introduce \textbf{M olGen}, a fragment-level, retrieval-augmented, two-stage framework for molecule generation under multi-property constraints. Stage I : Prototype generation: a multi-agent reasoner performs retrieval-anchored, fragment-level edits to produce a candidate near the feasible region. Stage II : RL-based fine-grained optimization: a fragment-level optimizer trained with Group Relative Policy Optimization (GRPO) applies one- or multi-hop refinements to explicitly minimize the property errors toward our target while regulating edit complexity and deviation from the prototype. A large, automatically curated dataset with reasoning chains of fragment edits and measured property deltas underpins both stages, enabling deterministic, reproducible supervision and controllable multi-hop reasoning. Unlike prior work, our framework better reasons about molecules by leveraging fragments and supports controllable refinement toward numeric targets. Experiments on generation under two sets of property constraints (QED, LogP, Molecular Weight and HOMO, LUMO) show consistent gains in validity and precise satisfaction of multi-property targets, outperforming strong LLMs and graph-based algorithms.
In recent years, with the rapid advancement of large language models (LLMs), role-playing language agents (RPLAs) have emerged as a prominent research focus at the intersection of natural language processing (NLP) and human-computer interaction. This paper systematically reviews the current development and key technologies of RPLAs, delineating the technological evolution from early rule-based template paradigms, through the language style imitation stage, to the cognitive simulation stage centered on personality modeling and memory mechanisms. It summarizes the critical technical pathways supporting high-quality role-playing, including psychological scale-driven character modeling, memory-augmented prompting mechanisms, and motivation-situation-based behavioral decision control. At the data level, the paper further analyzes the methods and challenges of constructing role-specific corpora, focusing on data sources, copyright constraints, and structured annotation processes. In terms of evaluation, it collates multi-dimensional assessment frameworks and benchmark datasets covering role knowledge, personality fidelity, value alignment, and interactive hallucination, while commenting on the advantages and disadvantages of methods such as human evaluation, reward models, and LLM-based scoring. Finally, the paper outlines future development directions of role-playing agents, including personality evolution modeling, multi-agent collaborative narrative, multimodal immersive interaction, and integration with cognitive neuroscience, aiming to provide a systematic perspective and methodological insights for subsequent research.
Optimizing communication topology in LLM-based multi-agent system is critical for enabling collective intelligence. Existing methods mainly rely on spatio-temporal interaction paradigms, where the sequential execution of multi-round dialogues incurs high latency and computation. Motivated by the recent insights that evaluation and debate mechanisms can improve problem-solving in multi-agent systems, we propose TopoDIM, a framework for one-shot Topology generation with Diverse Interaction Modes. Designed for decentralized execution to enhance adaptability and privacy, TopoDIM enables agents to autonomously construct heterogeneous communication without iterative coordination, achieving token efficiency and improved task performance. Experiments demonstrate that TopoDIM reduces total token consumption by 46.41% while improving average performance by 1.50% over state-of-the-art methods. Moreover, the framework exhibits strong adaptability in organizing communication among heterogeneous agents. Code is available at: https://anonymous.4open.science/r/TopoDIM-8D35/
Repository aware coding agents often struggle to recover build and test structure, especially in multilingual projects where cross language dependencies are encoded across heterogeneous build systems and tooling. We introduce the Repository Intelligence Graph (RIG), a deterministic, evidence backed architectural map that represents buildable components, aggregators, runners, tests, external packages, and package managers, connected by explicit dependency and coverage edges that trace back to concrete build and test definitions. We also present SPADE, a deterministic extractor that constructs RIG from build and test artifacts (currently with an automatic CMake plugin based on the CMake File API and CTest metadata), and exposes RIG as an LLM friendly JSON view that agents can treat as the authoritative description of repository structure. We evaluate three commercial agents (Claude Code, Cursor, Codex) on eight repositories spanning low to high build oriented complexity, including the real world MetaFFI project. Each agent answers thirty structured questions per repository with and without RIG in context, and we measure accuracy, wall clock completion time, and efficiency (seconds per correct answer). Across repositories and agents, providing RIG improves mean accuracy by 12.2\% and reduces completion time by 53.9\%, yielding a mean 57.8\% reduction in seconds per correct answer. Gains are larger in multilingual repositories, which improve by 17.7\% in accuracy and 69.5\% in efficiency on average, compared to 6.6\% and 46.1\% in single language repositories. Qualitative analysis suggests that RIG shifts failures from structural misunderstandings toward reasoning mistakes over a correct structure, while rare regressions highlight that graph based reasoning quality remains a key factor.
Large language models are increasingly deployed in multi-agent systems for strategic tasks, yet how design choices such as role-based personas and payoff visibility affect reasoning remains poorly understood. We investigate whether multi-agent systems function as strategic reasoners capable of payoff optimization or as identity-driven actors that prioritize role alignment over explicit incentives. Using Nash equilibrium achievement as a diagnostic for strategic reasoning, we conduct systematic experiments across four LLM architectures (Qwen-7B, Qwen-32B, Llama-8B, Mistral-7B) in complex environmental decision-making games involving four agents. We show that role identity bias fundamentally alters strategic reasoning even when payoff-optimal equilibria exist and complete payoff information is available. Removing personas and providing explicit payoffs enables Qwen models to achieve high Nash equilibrium rates, indicating that both conditions are necessary for strategic reasoning. In contrast, personas systematically bias equilibrium selection toward socially preferred outcomes: with personas present, all of the achieved equilibria correspond to Green Transition, while models entirely fail to reach equilibrium when Tragedy of the Commons is payoff-optimal. The effect of explicit payoffs depends entirely on persona presence, revealing strong interactions between representational design choices. We also observe clear model-dependent patterns. Qwen architectures are highly sensitive to both personas and payoff visibility, whereas Llama and Mistral exhibit rigid reasoning behavior across conditions. These findings demonstrate that representational choices are substantive governance decisions that determine whether multi-agent systems act as strategic reasoners or identity-driven actors, with important implications for real-world deployment.
The past year has marked a turning point in the evolution and real-world use of large language models (LLMs). With the release of the first widely adopted reasoning model, o1, on December 5th, 2024, the field shifted from single-pass pattern generation to multi-step deliberation inference, accelerating deployment, experimentation, and new classes of applications. As this shift unfolded at a rapid pace, our empirical understanding of how these models have actually been used in practice has lagged behind. In this work, we leverage the OpenRouter platform, which is an AI inference provider across a wide variety of LLMs, to analyze over 100 trillion tokens of real-world LLM interactions across tasks, geographies, and time. In our empirical study, we observe substantial adoption of open-weight models, the outsized popularity of creative roleplay (beyond just the productivity tasks many assume dominate) and coding assistance categories, plus the rise of agentic inference. Furthermore, our retention analysis identifies foundational cohorts: early users whose engagement persists far longer than later cohorts. We term this phenomenon the Cinderella "Glass Slipper" effect. These findings underscore that the way developers and end-users engage with LLMs "in the wild" is complex and multifaceted. We discuss implications for model builders, AI developers, and infrastructure providers, and outline how a data-driven understanding of usage can inform better design and deployment of LLM systems.
Large Language Models (LLMs) are increasingly used in mental health-related settings, yet they struggle to sustain realistic, goal-directed dialogue over extended interactions. While LLMs generate fluent responses, they optimize locally for the next turn rather than maintaining a coherent model of therapeutic progress, leading to brittleness and long-horizon drift. We introduce CALM-IT, a framework for generating and evaluating long-form Motivational Interviewing (MI) dialogues that explicitly models dual-actor conversational dynamics. CALM-IT represents therapist-client interaction as a bidirectional state-space process, in which both agents continuously update inferred alignment, mental states, and short-term goals to guide strategy selection and utterance generation. Across large-scale evaluations, CALM-IT consistently outperforms strong baselines in Effectiveness and Goal Alignment and remains substantially more stable as conversation length increases. Although CALM-IT initiates fewer therapist redirections, it achieves the highest client acceptance rate (64.3%), indicating more precise and therapeutically aligned intervention timing. Overall, CALM-IT provides evidence for modeling evolving conversational state being essential for generating high-quality long-form synthetic conversations.
Large Language Models (LLMs) are increasingly shaping human-computer interaction (HCI), from personalized assistants to social simulations. Beyond language competence, researchers are exploring whether LLMs can exhibit human-like characteristics that influence engagement, decision-making, and perceived realism. Personality, in particular, is critical, yet existing approaches often struggle to achieve both nuanced and adaptable expression. We present a framework that models LLM personality via Jungian psychological types, integrating three mechanisms: a dominant-auxiliary coordination mechanism for coherent core expression, a reinforcement-compensation mechanism for temporary adaptation to context, and a reflection mechanism that drives long-term personality evolution. This design allows the agent to maintain nuanced traits while dynamically adjusting to interaction demands and gradually updating its underlying structure. Personality alignment is evaluated using Myers-Briggs Type Indicator questionnaires and tested under diverse challenge scenarios as a preliminary structured assessment. Findings suggest that evolving, personality-aware LLMs can support coherent, context-sensitive interactions, enabling naturalistic agent design in HCI.
In this work we explore the performance and behavior of reasoning large language models to autonomously optimize atomic layer deposition (ALD) processes. In the ALD process optimization task, an agent built on top of a reasoning LLM has to find optimal dose times for an ALD precursor and a coreactant without any prior knowledge on the process, including whether it is actually self-limited. The agent is meant to interact iteratively with an ALD reactor in a fully unsupervised way. We evaluate this agent using a simple model of an ALD tool that incorporates ALD processes with different self-limited surface reaction pathways as well as a non self-limited component. Our results show that agents based on reasoning models like OpenAI's o3 and GPT5 consistently succeeded at completing this optimization task. However, we observed significant run-to-run variability due to the non deterministic nature of the model's response. In order to understand the logic followed by the reasoning model, the agent uses a two step process in which the model first generates an open response detailing the reasoning process. This response is then transformed into a structured output. An analysis of these reasoning traces showed that the logic of the model was sound and that its reasoning was based on the notions of self-limited process and saturation expected in the case of ALD. However, the agent can sometimes be misled by its own prior choices when exploring the optimization space.
The diversification of information access systems, from RAG to autonomous agents, creates a critical need for comparative user studies. However, the technical overhead to deploy and manage these distinct systems is a major barrier. We present UXLab, an open-source system for web-based user studies that addresses this challenge. Its core is a web-based dashboard enabling the complete, no-code configuration of complex experimental designs. Researchers can visually manage the full study, from recruitment to comparing backends like traditional search, vector databases, and LLMs. We demonstrate UXLab's value via a micro case study comparing user behavior with RAG versus an autonomous agent. UXLab allows researchers to focus on experimental design and analysis, supporting future multi-modal interaction research.
Most language-based assistants follow a reactive ask-and-respond paradigm, requiring users to explicitly state their needs. As a result, relevant but unexpressed needs often go unmet. Existing proactive agents attempt to address this gap either by eliciting further clarification, preserving this burden, or by extrapolating future needs from context, often leading to unnecessary or mistimed interventions. We introduce ProPer, Proactivity-driven Personalized agents, a novel two-agent architecture consisting of a Dimension Generating Agent (DGA) and a Response Generating Agent (RGA). DGA, a fine-tuned LLM agent, leverages explicit user data to generate multiple implicit dimensions (latent aspects relevant to the user's task but not considered by the user) or knowledge gaps. These dimensions are selectively filtered using a reranker based on quality, diversity, and task relevance. RGA then balances explicit and implicit dimensions to tailor personalized responses with timely and proactive interventions. We evaluate ProPer across multiple domains using a structured, gap-aware rubric that measures coverage, initiative appropriateness, and intent alignment. Our results show that ProPer improves quality scores and win rates across all domains, achieving up to 84% gains in single-turn evaluation and consistent dominance in multi-turn interactions.
Retrieval-augmented generation (RAG) has become the default strategy for providing large language model (LLM) agents with contextual knowledge. Yet RAG treats memory as a stateless lookup table: information persists indefinitely, retrieval is read-only, and temporal continuity is absent. We define the \textit{Continuum Memory Architecture} (CMA), a class of systems that maintain and update internal state across interactions through persistent storage, selective retention, associative routing, temporal chaining, and consolidation into higher-order abstractions. Rather than disclosing implementation specifics, we specify the architectural requirements CMA imposes and show consistent behavioral advantages on tasks that expose RAG's structural inability to accumulate, mutate, or disambiguate memory. The empirical probes (knowledge updates, temporal association, associative recall, contextual disambiguation) demonstrate that CMA is a necessary architectural primitive for long-horizon agents while highlighting open challenges around latency, drift, and interpretability.
Most existing Large Language Model (LLM)-based Multi-Agent Systems (MAS) rely on predefined workflows, where human engineers enumerate task states in advance and specify routing rules and contextual injections accordingly. Such workflow-driven designs are essentially rule-based decision trees, which suffer from two fundamental limitations: they require substantial manual effort to anticipate and encode possible task states, and they cannot exhaustively cover the state space of complex real-world tasks. To address these issues, we propose an Information-Flow-Orchestrated Multi-Agent Paradigm via Agent-to-Agent (A2A) Communication from CORAL, in which a dedicated information flow orchestrator continuously monitors task progress and dynamically coordinates other agents through the A2A toolkit using natural language, without relying on predefined workflows. We evaluate our approach on the general-purpose benchmark GAIA, using the representative workflow-based MAS OWL as the baseline while controlling for agent roles and underlying models. Under the pass@1 setting, our method achieves 63.64% accuracy, outperforming OWL's 55.15% by 8.49 percentage points with comparable token consumption. Further case-level analysis shows that our paradigm enables more flexible task monitoring and more robust handling of edge cases. Our implementation is publicly available at: https://github.com/Coral-Protocol/Beyond-Rule-Based-Workflows