Aligning large language models (LLMs) with human preferences is critical for enhancing LLMs' safety, helpfulness, humor, faithfulness, etc. Current reinforcement learning from human feedback (RLHF) mainly focuses on a fixed reward learned from average human ratings, which may weaken the adaptability and controllability of varying preferences. However, creating personalized LLMs requires aligning LLMs with individual human preferences, which is non-trivial due to the scarce data per user and the diversity of user preferences in multi-objective trade-offs, varying from emphasizing empathy in certain contexts to demanding efficiency and precision in others. Can we train one LLM to produce personalized outputs across different user preferences on the Pareto front? In this paper, we introduce Multi-Objective Control (MOC), which trains a single LLM to directly generate responses in the preference-defined regions of the Pareto front. Our approach introduces multi-objective optimization (MOO) principles into RLHF to train an LLM as a preference-conditioned policy network. We improve the computational efficiency of MOC by applying MOO at the policy level, enabling us to fine-tune a 7B-parameter model on a single A6000 GPU. Extensive experiments demonstrate the advantages of MOC over baselines in three aspects: (i) controllability of LLM outputs w.r.t. user preferences on the trade-off among multiple rewards; (ii) quality and diversity of LLM outputs, measured by the hyper-volume of multiple solutions achieved; and (iii) generalization to unseen preferences. These results highlight MOC's potential for real-world applications requiring scalable and customizable LLMs.

Large Language Models (LLMs) have achieved human-level performance on a wide array of benchmarks, yet deploying them in critical applications requires that their internal confidence track actual competence. In this study, we evaluate the alignment between LLMs’ self-assessment and their actual competence in solving mathematical problems. We probe this gap on three mathematical datasets: MATH, Math500 and GSM8K, by eliciting LLM’s confidence and comparing it with its solution accuracy. Experiments on 8 open-source models from the Qwen and LLaMA families show three key findings: (1) Most of models display a certain degree of misalignment; (2) Scale and domain-specific fine-tuning matter: 7B-parameter and math-tuned Qwen variants narrow the confidence–performance gap, whereas similarly sized but untuned LLaMA models remain poorly consistent; (3) Misalignment is also sensitive to prompt design.

Recent advances in vision-language models have opened up new possibilities for reasoning-driven image geolocalization. However, existing approaches often rely on synthetic reasoning annotations or external image retrieval, which can limit interpretability and generalizability. In this paper, we present Geo-R, a retrieval-free framework that uncovers structured reasoning paths from existing ground-truth coordinates and optimizes geolocation accuracy via reinforcement learning. We propose the Chain of Region, a rule-based hierarchical reasoning paradigm that generates precise, interpretable supervision by mapping GPS coordinates to geographic entities (e.g., country, province, city) without relying on model-generated or synthetic labels. Building on this, we introduce a lightweight reinforcement learning strategy with coordinate-aligned rewards based on Haversine distance, enabling the model to refine predictions through spatially meaningful feedback. Our approach bridges structured geographic reasoning with direct spatial supervision, yielding improved localization accuracy, stronger generalization, and more transparent inference. Experimental results across multiple benchmarks confirm the effectiveness of Geo-R, establishing a new retrieval-free paradigm for scalable and interpretable image geolocalization. To facilitate further research and ensure reproducibility, both the model and code will be made publicly available.

Recent agentic search systems have made substantial progress by emphasising deep, multi-step reasoning. However, this focus often overlooks the challenges of wide-scale information synthesis, where agents must aggregate large volumes of heterogeneous evidence across many sources. As a result, most existing large language model agent systems face severe limitations in data-intensive settings, including context saturation, cascading error propagation, and high end-to-end latency. To address these challenges, we present , a hierarchical framework based on principle of near-decomposability, containing a strategic Host, multiple Managers and parallel Workers. By leveraging aggregation and reflection mechanisms at the Manager layer, our framework enforces strict context isolation to prevent saturation and error propagation. Simultaneously, the parallelism in worker layer accelerates the speed of overall task execution, mitigating the significant latency. Our evaluation on two complementary benchmarks demonstrates both efficiency (3-5 × speed-up) and effectiveness, achieving a 8.4% success rate on WideSearch-en and 52.9% accuracy on BrowseComp-zh. The code is released at https://github.com/agent-on-the-fly/InfoSeeker

With the rapid advancement of Large Language Models (LLMs), the safety of LLMs has been a critical concern requiring precise assessment. Current benchmarks primarily concentrate on single-turn dialogues or a single jailbreak attack method to assess the safety. Additionally, these benchmarks have not taken into account the LLM's capability of identifying and handling unsafe information in detail. To address these issues, we propose a fine-grained benchmark SafeDialBench for evaluating the safety of LLMs across various jailbreak attacks in multi-turn dialogues. Specifically, we design a two-tier hierarchical safety taxonomy that considers 6 safety dimensions and generates more than 4000 multi-turn dialogues in both Chinese and English under 22 dialogue scenarios. We employ 7 jailbreak attack strategies, such as reference attack and purpose reverse, to enhance the dataset quality for dialogue generation. Notably, we construct an innovative assessment framework of LLMs, measuring capabilities in detecting, and handling unsafe information and maintaining consistency when facing jailbreak attacks. Experimental results across 17 LLMs reveal that Yi-34B-Chat and GLM4-9B-Chat demonstrate superior safety performance, while Llama3.1-8B-Instruct and o3-mini exhibit safety vulnerabilities.

The role of natural language constraints in Safe Multi-agent Reinforcement Learning (MARL) is crucial, yet often overlooked. While Safe MARL has vast potential, especially in fields like robotics and autonomous vehicles, its full potential is limited by the need to define constraints in pre-designed mathematical terms, which requires extensive domain expertise and reinforcement learning knowledge, hindering its broader adoption. To address this limitation and make Safe MARL more accessible and adaptable, we propose a novel approach named Safe Multi-agent Reinforcement Learning with Natural Language constraints (SMALL). Our method leverages fine-tuned language models to interpret and process free-form textual constraints, converting them into semantic embeddings that capture the essence of prohibited states and behaviours. These embeddings are then integrated into the multi-agent policy learning process, enabling agents to learn policies that minimize constraint violations while optimizing rewards. To evaluate the effectiveness of SMALL, we introduce the LaMaSafe, a multi-task benchmark designed to assess the performance of multiple agents in adhering to natural language constraints. Empirical evaluations across various environments demonstrate that SMALL achieves comparable rewards and significantly fewer constraint violations, highlighting its effectiveness in understanding and enforcing natural language constraints.

Large language models (LLMs) have achieved remarkable success in many domains, but concerns about data quality and privacy are growing. Federated Learning (FL) offers a privacy-preserving solution by training a model on local clients without sharing data. However, the impact of biased private data on LLMs fine-tuned through FL remains understudied. This work investigates how client-side biased data affects the global model during federated fine-tuning of LLMs. We simulate realistic scenarios where some clients possess datasets containing social biases (stereotypes, discriminatory language) while others have clean data through extensive experiments with popular FL algorithms (FedAvg, FedAdam and FedProx) and popular LLMs (LLaMA, Mistral, Phi-3 and Gemma) across datasets with varying bias proportions (33%, 66%, 100%). Our findings reveal that 1) FedAdam consistently shows the lowest bias propagation, reducing CrowS-Pairs scores by up to 15% compared to FedAvg; 2) Even small amounts of biased data (33%) can significantly influence global model bias; 3) Mixed biased and neutral data distributions lead to 5%-7% higher bias scores than segregated distributions. Additionally, we propose Bias-Aware Model Aggregation (BAMA), a novel debiasing method for federated fine-tuning that consistently reduces bias across various models and algorithms.

Large Language Models (LLMs) have recently emerged as powerful controllers for interactive agents in complex environments, yet training them to perform reliable long-horizon decision making remains a fundamental challenge. A key difficulty lies in credit assignment: agents often receive delayed rewards only at the end of episodes. In this paper, we propose Q-Evolve, a self-evolving framework for LLM agents that unifies automatic process-reward labeling and policy learning within a principled in-distribution reinforcement learning paradigm. In each evolving iteration, our method learns an in-distribution critic from a hybrid off-policy dataset that combines expert demonstrations with agent-generated trajectories, stabilizing Bellman backups in sparse-reward settings via a weighted Implicit Q-Learning objective. The learned value function is then used to derive step-wise process rewards through advantage estimation, enabling dense and reliable supervision without environment backtracking or human annotation. Leveraging these signals, we perform behavior-proximal policy optimization that evolves the agent over the data used for process reward labeling, allowing iterative self-improvement without exacerbating distribution shift. We evaluate our method on AlfWorld, WebShop, and ScienceWorld, showing Q-Evolve outperforms strong baselines in sample efficiency, robustness, and overall task performance. Our results demonstrate that stable agent self-evolution is achievable through the co-evolution of process-level supervision and policy, both grounded within a shared in-distribution learning loop.

Exploration is critical for cooperative multi agent reinforcement learning (MARL) to improve sample efficiency. However, existing intrinsic motivation based exploration strategies in MARL overlook the causal relationships among agents, global states, and rewards, suffering from interference by irrelevant factors and resulting in sample inefficiency. To address this issue, we propose Causality aware Efficient Exploration (CEE), a novel framework that enhances sample efficiency by inferring causal relationships between agents, global states with respect to rewards, thereby enabling causality guided exploration. Specifically, CEE operates through two components. First, CEE identifies causal relationships between global states and rewards, filtering out causally irrelevant state features that do not have a high impact on rewards to keep decision critical state information. Second, CEE discovers causal relationships between agents' behaviors and rewards to quantify each agent's contribution to collective performance. To achieve this, we introduce a causal entropy objective that promotes exploration aligned with decision critical aspects of the underlying causal structure. We provide comprehensive validation through experiments on 21 challenging tasks spanning SMAC, SMAC v2, and Google Research Football (GRF) environments. Our results demonstrate that CEE achieves superior performance in terms of sample efficiency and asymptotic performance compared to existing MARL methods.

Multilingual large language models (LLMs) are advancing rapidly, with new models frequently claiming support for an increasing number of languages. However, existing evaluation datasets are limited and lack cross-lingual alignment, leaving assessments of multilingual capabilities fragmented in both language and skill coverage. To address this, we introduce MuBench, a benchmark covering 61 languages and evaluating a broad range of capabilities. We evaluate several state-of-the-art multilingual LLMs and find notable gaps between claimed and actual language coverage, particularly a persistent performance disparity between English and low-resource languages. Leveraging MuBench's alignment, we propose Multilingual Consistency (MLC) as a complementary metric to accuracy for analyzing performance bottlenecks and guiding model improvement. Finally, we pretrain a suite of 1.2B-parameter models on English and Chinese with 500B tokens, varying language ratios and parallel data proportions to investigate cross-lingual transfer dynamics.

High-quality and diverse multimodal data are essential for improving vision-language models (VLMs), yet existing datasets often contain noisy, redundant, and poorly aligned samples. To address these problems, data filtering is commonly used to enhance the efficiency and performance of multimodal learning, but it introduces extra computational cost because filtering models are usually trained on the same data they are meant to screen. To reduce this cost, we study DOSE, which explores whether off-the-shelf pretrained models that have never seen the target data can be used to select training samples for larger and stronger multimodal models without any task-specific training. Even without fine-tuning, these models can effectively assess text quality and image-text alignment to guide data selection. Based on this, we build a joint quality-alignment distribution and apply adaptive weighted sampling to select informative samples while maintaining long-tail diversity. This approach enhances data diversity, enabling models trained on DOSE-filtered data to match or surpass those trained on the full dataset on standard VQA and math benchmarks. Extensive experiments demonstrate its effectiveness, efficiency, and scalability.

Benchmarks play a central role in reinforcement learning (RL) research, yet their computational constraints often shape what is studied. Despite the motivation of lifelong learning, most continual RL papers consider only 3–10 sequential tasks, as CPU-bound environments make longer sequences impractical. Meanwhile, continual learning in cooperative multi-agent settings remains largely unexplored. To address these gaps, we introduce (ulti-agent nvironments for daptive earning), the first benchmark for continual multi-agent RL. By leveraging JAX and GPU acceleration, MEAL enables training on sequences of 100 tasks on a single GPU in a few hours. We find that long task sequences reveal failure modes that do not appear at smaller scales.

Analyzing cultural-heritage artifacts remains challenging for MLLMs: general models lack domain expertise, and SFT often overfits superficial patterns, yielding brittle reasoning for authentication and historical attribution. This raises the question of how to equip MLLMs with robust, expert-level reasoning for ancient Greek pottery. We present VaseVL, an SFT-then-RL system that turns evaluation into supervision: we construct a taxonomy of question types, probe the SFT model to localize type-specific performance gaps, and optimize with type-conditioned, compositionality-oriented rewards targeting those gaps. We also release VaseVQA, a comprehensive benchmark of 31,773 images designed to probe deep understanding. Experiments show state-of-the-art results on style classification and historical attribution with marked gains in compositional robustness over SFT-only baselines, validating diagnosis-guided, taxonomy-conditioned reward engineering and providing a reusable resource for future research. Code and dataset will be available at https://github.com/AIGeeksGroup/VaseVQA.

Recent advances in visual language models have enabled autonomous agents for complex reasoning, tool use, and document understanding. However, existing document agents mainly transform papers into static artifacts such as summaries, webpages, or slides, which are insufficient for technical papers involving dynamic mechanisms and state transitions. In this work, we propose a Paper-to-Interactive-System Agent that converts research papers into executable interactive web systems. Given a PDF paper, the agent performs end-to-end processing without human intervention, including paper understanding, system modeling, and interactive webpage synthesis, enabling users to manipulate inputs and observe dynamic behaviors. To evaluate this task, we introduce a benchmark of 19 research papers paired with expert-built interactive systems as ground truth. We further propose PaperVoyager, a structured generation framework that explicitly models mechanisms and interaction logic during synthesis. Experiments show that PaperVoyager significantly improves the quality of generated interactive systems, offering a new paradigm for interactive scientific paper understanding.

Existing retrieval-augmented generation (RAG) methods often treat retrieval as a one-off operation, yet recent work suggests that iteratively refining the retrieval step can yield substantial gains in relevance and downstream generation quality. However, prior iterative-retrieval approaches typically optimize only the retriever’s ranking function or only post-hoc document refinement, and they require expensive retriever retraining or complex multi-stage pipelines. To address these challenges, we propose Adaptive Iterative Retrieval for Retrieval-Augmented Generation (AIR-RAG), an adaptive, iterative retrieval framework designed to optimize both document relevance and LLM alignment within the RAG pipeline. By leveraging adaptive feedback, AIR-RAG simultaneously enhances retrieval ranking and document refinement across multiple iterations, eliminating the need for complex retraining pipelines and enabling seamless integration with existing systems. In extensive evaluations against state-of-the-art RAG methods across several benchmark datasets including TriviaQA, PopQA, HotpotQA, WikiMultiHop, PubHealth, and StrategyQA, AIR-RAG consistently demonstrates superior performance, underscoring its effectiveness in enhancing retrieval-augmented generation systems. Our code and data are available anonymously at https://github.com/aialt/AIR-RAG.

Multi-agent LLM systems commonly use an orchestrator to decompose a task for a team of workers and then improve through textual reflection. Despite strong empirical results, these systems lack a unified account of coordination, memory improvement, and the role of external verification. We model orchestrator-worker interaction as a bilevel coordination game: under bounded coupling, the workers' local-update game is an approximate potential game whose equilibrium slack is controlled by decomposition quality. We then analyse reflection as stochastic movement over semantic memory states. For free-form reflection, we derive a finite-time upper bound, prove worst-case tightness, and give a positive lower bound under a falsifiable persistent-harm condition. We further prove an information-theoretic impossibility result: no gate that observes only the generated transcript can improve uniformly over text-indistinguishable environments, whereas an environment-grounded gate can. Motivated by this separation, we introduce Stochastic Reflective Memory Ascent (SRMA), which accepts a candidate memory only after a grounded evaluation risk strictly decreases. Under calibration and non-degenerate corrective mass, SRMA converges exactly, geometrically or polynomially; matching constructions show that both rate regimes are order-tight. We also provide confidence gating for stochastic evaluation and re-anchoring guarantees for piecewise-stationary environments. Experiments instantiate these objects with environment-grounded metrics and test the predicted coordination and drift laws. On 500 SWE-bench instances, the complete Kimi-based system resolves 72.2

Online rating systems are an indispensable building block for many web applications, yet they are pervasively affected by data biases that distort our understanding of user preferences. A key source of this bias is the “Spiral of Silence” (SoS) phenomenon, where users who perceive their opinions to be in the minority tend to withhold them, leading to non-random missing data. While initial research has begun to explore this issue, existing SoS models for recommender systems fail to capture real-world complexities, such as rating recency, user heterogeneity (e.g., “hardcore” users immune to social pressure), and asymmetric responses to conforming versus dissenting opinions. Furthermore, these models often lack formal theoretical guarantees. To fill this gap, we introduce Bi-MCP, a modular probabilistic framework to quantify and mitigate SoS effects. Bi-MCP models a bifurcated user population (hardcore vs. conformist) and their bidirectional (asymmetric) response to the perceived opinion climate. The framework is (i) Expressive, capturing novel factors missed by prior work; (ii) Modular, allowing its components to be extended or replaced to support future research; and (iii) Theoretically grounded, with a formal proof of convergence for its Generalized Expectation-Maximization (GEM) inference algorithm. Experimental results on four large-scale public datasets demonstrate that our model significantly improves recommendation accuracy over state-of-the-art baselines, showcasing the practical benefit of explicitly modeling these nuanced SoS effects.

As autonomous web agents are increasingly deployed to perform real-world tasks, ensuring their safety has become a critical concern. In this work, we study web agent behavior under realistic deceptive interfaces in the e-commerce domain. We introduce WebDecept, a lightweight and configurable plugin framework that enables controlled injection of deceptive interface patterns into existing web environments. Using WebDecept, we instantiate seven deceptive patterns commonly observed on the open web, including targeted advertisements, domain redirection, and shopping manipulation. By injecting these patterns into the frontend during task execution, we perform controlled evaluation of multiple multimodal web agents. Our results show that current web agents are highly susceptible to multiple classes of deceptive interfaces, and that prompt-based constraints are often insufficient to mitigate these failures. We further analyze how the design choices of deceptive patterns influence the success of such manipulations. These findings highlight safety challenges that should be addressed as web agents are scaled toward real-world deployment.

Individual agent capabilities have advanced rapidly through modular skills and tool integrations, yet multi-agent systems remain constrained by fixed team structures, tightly coupled coordination logic, and session-bound learning. We argue that this reflects a deeper absence: a principled organisational layer that governs how a workforce of agents is assembled, governed, and improved over time, decoupled from what individual agents know. To fill this gap, we introduce OneManCompany (OMC), a framework that elevates multi-agent systems to the organisational level. OMC encapsulates skills, tools, and runtime configurations into portable agent identities called Talents, orchestrated through typed organisational interfaces that abstract over heterogeneous backends. A community-driven Talent Market enables on-demand recruitment, allowing the organisation to close capability gaps and reconfigure itself dynamically during execution. Organisational decision-making is operationalised through an Explore-Execute-Review (E^2R) tree search, which unifies planning, execution, and evaluation in a single hierarchical loop: tasks are decomposed top-down into accountable units and execution outcomes are aggregated bottom-up to drive systematic review and refinement. This loop provides formal guarantees on termination and deadlock freedom while mirroring the feedback mechanisms of human enterprises. Together, these contributions transform multi-agent systems from static, pre-configured pipelines into self-organising and self-improving AI organisations capable of adapting to open-ended tasks across diverse domains. Empirical evaluation on PRDBench shows that OMC achieves an 84.67% success rate, surpassing the state of the art by 15.48 percentage points, with cross-domain case studies further demonstrating its generality.