Large Language Models (LLMs) are frequently portrayed as general-purpose solvers capable of solving arbitrary tasks. We argue that this view overlooks a fundamental constraint: language is a compressed and capacity-limited interface for conveying task information. Modelling User–System interaction as a bilevel cheap-talk game, we analyse how latent tasks are encoded into prompts and reinterpreted under alignment and safety constraints. We introduce a conceptual decomposition separating task inference from execution and derive PAC-Bayes bounds that distinguish finite-sample estimation error from irreducible structural limitations. Our first main result establishes an expressivity floor: language acts as a capacity-limited communication channel, and whenever the informational complexity of a task family exceeds the capacity of that channel, distinct tasks become unavoidably indistinguishable to the Solver, inducing a strictly positive error floor that cannot be eliminated by additional data, optimisation, or model scaling alone. We then establish an objective-misalignment floor: when alignment constraints restrict the admissible output set, the User-ideal distribution may lie outside the feasible class, inducing an irreducible distortion. Together, these results yield a formal negative conclusion: prompt-conditioned LLMs are not universal problem solvers through prompting alone, as there exist task families for which correct behaviour is provably unattainable even in the infinite-data regime. More broadly, our analysis shows the limits of prompt-based generalisation arise from information-constrained communication and alignment-constrained objectives. This suggests that interfaces beyond natural language, including multimodal observations and, external memory, may reduce the inherent LLM limitations by increasing the task-relevant information available to the System.

Massive Multiple-Input Multiple-Output (MIMO) is a key technology in 5G/Beyond 5G networks, and is expected to play an even more crucial role in 6G networks. It faces challenges in enhancing performance while managing interference and reducing computational complexity in scale-expanded multi-user MIMO scheduling. The sharing of channel state information (CSI) for inter-cell coordination, while essential, increases the signaling overhead and delay in multi-cell scenarios. A wide range of deep learning and reinforcement learning methods has been introduced to tackle the aforementioned challenges. However, in the multi-cell environments, the dimensionality of channel state information (CSI) as well as the size of user combination sets can grow at an exponential rate. This phenomenon significantly amplifies the complexity of neural networks and induces severe convergence challenges. To address the single-cell downlink multi-user scheduling problem under such circumstances, this paper introduces a novel Pointer Network-based Actor-Critic reinforcement learning algorithm (PNAC). Within the PNAC framework, the pointer network is employed as the policy network. This approach elegantly reframes the complex combinatorial user selection problem as a sequential decision making process governed by conditional probabilities. This transformation effectively mitigates the high-dimensionality issue inherent in the action space, thereby enhancing the algorithm's efficiency and applicability in complex massive MIMO environments. In multi-cell scenarios, building upon PNAC, we propose a more robust multi-agent reinforcement learning algorithm, MA-PNAC, which follows the centralized training with decentralized execution (CTDE) framework to outcomes demonstrate that the system spectral efficiency achieved by our two algorithms closely approaches that of the widely used greedy algorithms in the industry, while significantly reducing computational complexity in both single-cell and multi-cell scenarios. Our methods reduce running time by one to two orders of magnitude compared to greedy algorithms. Moreover, the multi-cell scheduling in our approach eliminates the need for signal exchange among base stations in Time Division Duplex (TDD) systems in inference phase, thus avoiding signaling overhead and delay.

Large language models (LLMs) have become a central foundation of modern artificial intelligence, yet their lifecycle remains constrained by a rigid separation between training and deployment, after which learning effectively ceases. This limitation contrasts with natural intelligence, which continually adapts through interaction with its environment. In this paper, we formalise deployment-time learning (DTL) as the third stage in the LLM lifecycle that enables LLM agents to improve from experience during deployment without modifying model parameters. We present CASCADE (CASe-based Continual Adaptation during DEployment), a general and principled framework that equips LLM agents with an explicit, evolving episodic memory. CASCADE formulates experience reuse as a contextual bandit problem, enabling principled exploration-exploitation trade-offs and establishing no-regret guarantees over long-term interactions. This design allows agents to accumulate, select, and refine task-relevant cases, transforming past experience into actionable knowledge. Across 16 diverse tasks spanning medical diagnosis, legal analysis, code generation, web search, tool use, and embodied interaction, CASCADE improves macro-averaged success rate by 20.9

Due to the distributed devices in Mobile Edge Computing (MEC) systems, traditional machine learning is not suitable, and federated learning is typically considered instead. In federated learning, there are two critical challenges: 1) the data on distributed learners is heterogeneous, and 2) communication resources within the network are limited. In this work, we propose a framework, Federated Dissimilarity Measure (FedDM), which can be regarded as an adaptively enhanced version of the Federated Proximal (FedProx) algorithm. This adaptiveness is primarily manifested in two aspects: (i) how it adaptively adjusts the proximity between the local models on different learners and the global model; and (ii) how it adaptively aggregates local model parameters. Building on the FedProx model, FedDM incorporates the concept of the Lagrangian multiplier to control the proximal coefficients of different learners, using “parameter dissimilarity" to address data heterogeneity. It explicitly captures the essence of using “loss dissimilarity" to adaptively adjust the aggregation frequency on distributed learners, thereby reducing communication overhead. Theoretically, we provide the performance upper bounds and convergence analysis of our proposed FedDM. Experiment results demonstrate that FedDM allows for higher accuracy and lower communication overhead compared to the baselines across a suite of realistic datasets.

This paper presents a novel approach to reinforcement learning (RL) for control systems that provides probabilistic stability guarantees using finite data. Leveraging Lyapunov's method, we propose a probabilistic stability theorem that ensures mean square stability using only a finite number of sampled trajectories. The probability of stability increases with the number and length of trajectories, converging to certainty as data size grows. Additionally, we derive a policy gradient theorem for stabilizing policy learning and develop an RL algorithm, L-REINFORCE, that extends the classical REINFORCE algorithm to stabilization problems. The effectiveness of L-REINFORCE is demonstrated through simulations on a Cartpole task, where it outperforms the baseline in ensuring stability. This work bridges a critical gap between RL and control theory, enabling stability analysis and controller design in a model-free framework with finite data.

Is monolithic scaling the only path to AGI? This paper challenges the dogma that purely scaling a single model is enough to achieve universal super-intelligence. Instead, we identify Agentic AI as the necessary evolution for handling complex, real-world task distributions to achieve AGI in the human world. Through concrete theoretical derivations, we contrast the optimization constraints of monolithic learners against the efficiency of Agentic systems, evolving from simple routing mechanisms to general Directed Acyclic Graphs (DAGs) of Agents. We demonstrate that Agentic AI offers superior generalization and efficiency. Finally, we reinterpret the instability of current multi-agent frameworks and call for more future actions on Agentic AI.

LLM agents benefit from reusable skills, yet test-time tasks often require guidance more specific than a static skill library can provide. We propose SkillTTA, a Test-Time Adaptive Skill Synthesis method that retrieves a small set of training trajectories relevant to the current task and synthesizes them into a temporary, task-specific textual skill. The solver model is kept fixed, so adaptation happens entirely through generated context rather than parameter updates. We evaluate the method on SpreadsheetBench, ALFWorld, and BigCodeBench. Compared with static trajectory-to-skill synthesis using GPT-5.5, task-specific skills improve SpreadsheetBench Pass@1 from 0.397 to 0.505 and BigCodeBench Pass@1 from 0.517 to 0.651. On ALFWorld, the method matches a heavier memory-learning baseline within four points of success rate while producing the shortest successful trajectories among reported methods. Ablations on SpreadsheetBench further show that synthesized skills outperform raw trajectory prompting, that top-k retrieval should stay small, and that failed trajectories are especially useful because they expose recurring evaluator-facing mistakes.

Large language model (LLM) agents have achieved strong performance on a wide range of benchmarks, yet most evaluations assume static environments. In contrast, real-world deployment is inherently dynamic, requiring agents to continually align their knowledge, skills, and behavior with changing environments and updated task conditions. To address this gap, we introduce EvoArena, a benchmark suite that models environment changes as sequences of progressive updates across terminal, software, and social domains. We further propose EvoMem, a patch-based memory paradigm that records memory evolution as structured update histories, enabling agents to reason about environmental evolution through changes in their memory. Experiments show that current agents struggle on EvoArena, achieving an average accuracy of 39.6

Gaussian process (GP) regression with a single global GP (GP-glo) incurs cubic computational cost, limiting scalability to large datasets. Product-of-experts GP models (GP-pro), which combine local GP models to capture global correlations, alleviate this computational burden. However, training local experts on disjoint data subsets can lead to overestimated posterior variances. We propose GP-pro-c, a product-of-experts GP model that calibrates these variances using an information-based method. The method exploits the monotonicity and submodularity of information gain in GPs to define a calibration ratio that reduces the posterior variance of individual local GP models. We evaluate GP-pro-c using negative log-likelihood (NLL), root mean squared error (RMSE), and expected normalised calibration error (ENCE). Experiments on four synthetic functions and six regression datasets show that GP-pro-c achieves average reductions of 2.3

The aim of the knowledge graph-based question generation (KGQG) task is to generate an answerable, fluent question from a ternary knowledge graph and a target answer. Existing KGQG that study knowledge graph subgraphs and the question of target answer generation do not effectively capture the critical semantic information between tokens within nodes/edges subgraphs and fail to make full use of target answers and answer markers. This has led to the generation of disfluent and unanswerable questions. To address these problems, we propose a model called knowledge graph question generation based on crucial semantic information (KGQG-CSI). Our proposed model utilizes the critical semantic information encoding module to dynamically learn the degree of significance of tokens within the edges and nodes of fused answers, capturing critical semantic information that would remedy disfluency. In addition, the target answers and answer markers are sufficiently integrated with the nodes to make the generated questions answerable. First, the attention mechanism is used to allow the nodes to interact with the target answers, thereby expressing the semantic information related to the answers more accurately. The nodes that have been processed through the critical semantic information encoding module are then spliced with the answer markers to reduce the ambiguous information. The experimental results on two public datasets show that the results of the proposed model outperform the existing methods.

The goal of artificial intelligence is to create agents capable of general, adaptive behaviour in open-ended environments. Guided by the "Bitter Lesson", we argue that the most effective path toward this goal is to systematically remove human priors and allow intelligence to naturally emerge through interaction with a "Big World" that is orders of magnitude more complex than the agent itself. We propose the mobile Graphical User Interface (GUI) as a practical proxy for such a world and introduce Darwin Mobile Agent, an open-source infrastructure designed as a foundation for autonomous reinforcement learning in this domain. This framework addresses the data-collection bottleneck in real-world mobile interactions by using an asynchronous agent-environment loop across parallel cloud-phone instances. We further propose a conceptual roadmap to systematically remove human priors from three fundamental pillars of a self-evolving agent: task curricula, outcome verification, and memory management. We validate that the Darwin infrastructure provides the stability and scalability required for the first stage of this roadmap: policy optimisation in the GUI domain. This work establishes the practical and theoretical foundation necessary to move toward truly autonomous, self-evolving GUI agents.

As large language model (LLM) agents increasingly learn from experience, they primarily rely on trajectory-level reflection to extract insights. Viewed through the lens of predictive knowledge, we argue that this approach operates on episodic hindsight rather than predictive foresight, yielding brittle, path-dependent heuristics. To address this, we propose Stateful Knowledge Learning (SKL). SKL shifts the agent's focus from trajectory-level summarization to maintaining Stateful Knowledge: explicit, declarative predictive assessments anchored to state. We first demonstrate a motivating example showing how stateful knowledge provides granularity, enhances generalization, and enables knowledge bootstrapping. To further scale up the idea, we introduce two algorithms via self-distillation (SKL-SD) and reinforcement learning (SKL-RL), training agents to autonomously extract state-grounded predictive knowledge from experience and learn to leverage it for policy making. Experiments on interactive environments (WebShop, ScienceWorld) and a complex reasoning task (ChessPuzzles) demonstrate that equipping models with the inherent ability to learn stateful predictive knowledge significantly outpaces current reflection-based training paradigms.

Large language model (LLM) agent systems are increasingly expected to improve after deployment, but existing work often decouples two adaptation targets: skill evolution and multi-agent system (MAS) restructuring. This separation can create organization bottlenecks, context pressure, and mis-specialization. We present SkillMAS, a non-parametric framework for adaptive specialization in multi-agent systems that couples skill evolution with MAS restructuring. SkillMAS uses Utility Learning to assign credit from verified execution traces, bounded skill evolution to refine reusable procedures without unfiltered library growth, and evidence-gated MAS restructuring when retained failures and Executor Utility indicate a structural mismatch. Across embodied manipulation, command-line execution, and retail workflows, SkillMAS is competitive under the reported harnesses while clarifying how post-deployment specialization is attributed, updated, and applied.

Large Language Models (LLMs) excel at various reasoning tasks but still encounter challenges such as hallucination and factual inconsistency in knowledge-intensive tasks, primarily due to a lack of external knowledge and factual verification. These challenges could be mitigated by leveraging knowledge graphs (KGs) to support more reliable LLM reasoning. However, existing KG-augmented LLM frameworks still rely on static integration mechanisms that cannot adjust reasoning in response to evolving context and retrieved evidence, resulting in error propagation and incomplete reasoning. To alleviate these issues, we propose **V**erify-**o**n-**G**raph (**VoG**), a scalable and model-agnostic framework to enhance LLM reasoning via iterative retrieval, stepwise verification, and adaptive revision. Besides performing KG retrieval guided by an initially generated reasoning plan, VoG iteratively verifies and revises the reasoning plan, correcting intermediate errors in consideration of the varying contextual conditions. During plan revision, VoG leverages a context-aware multi-armed bandit strategy, guided by reward signals that capture uncertainty and semantic consistency, to enhance the alignment between the reasoning plan and retrieved evidence in a more adaptive and reliable way. Experimental results across three benchmark datasets show that VoG consistently improves both reasoning accuracy and efficiency. Our code is available at https://anonymous.4open.science/r/VoG-132C/.

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

Autonomous robots capable of turning natural-language instructions into reliable physical actions remain a central challenge in artificial intelligence. Here we show that connecting a large language model agent to the robot operating system enables a versatile framework for embodied intelligence, and we release the complete implementation as freely available open-source code. The agent automatically translates large language model outputs into robot actions, supports interchangeable execution modes (inline code or behaviour trees), learns new atomic skills via imitation, and continually refines them through automated optimization and reflection from human or environmental feedback. Extensive experiments validate the framework, showcasing robustness, scalability and versatility in diverse scenarios and embodiments, including long-horizon tasks, tabletop rearrangements, dynamic task optimization and remote supervisory control. Moreover, all the results presented in this work were achieved by utilizing open-source pretrained large language models.

With the advancement of vision-language models, web automation has made significant progress. However, deploying autonomous agents in real-world settings remains challenging, primarily due to site heterogeneity, where generalist models lack domain-specific priors for diverse interfaces, and long-horizon instability, characterized by the accumulation of decision drift over extended interactions. To address these challenges, we introduce ColorBrowserAgent (Complex Long-Horizon Browser Agent), a knowledge-evolving agent for robust web automation. Our approach addresses these challenges through two synergistic mechanisms: human-in-the-loop knowledge adaptation that transforms sparse human feedback into reusable domain knowledge, and knowledge-aligned progressive summarization that stabilizes long interactions through memory compression. Extensive experiments on WebArena, WebChoreArena and industrial deployment show that ColorBrowserAgent consistently outperforms strong baselines. It achieves a state-of-the-art success rate of 71.2

App agents, which autonomously operate mobile Apps through Graphical User Interfaces (GUIs), have gained significant interest in real-world applications. Yet, they often struggle with long-horizon planning, failing to find the optimal actions for complex tasks with longer steps. To address this, world models are used to predict the next GUI observation based on user actions, enabling more effective agent planning. However, existing world models primarily focus on generating only textual descriptions, lacking essential visual details. To fill this gap, we propose ViMo, the first visual world model designed to generate future App observations as images. For the challenge of generating text in image patches, where even minor pixel errors can distort readability, we decompose GUI generation into graphic and text content generation. We propose a novel data representation, the Symbolic Text Representation (STR) to overlay text content with symbolic placeholders while preserving graphics. With this design, ViMo employs a STR Predictor to predict future GUIs' graphics and a GUI-text Predictor for generating the corresponding text. Moreover, we deploy ViMo to enhance agent-focused tasks by predicting the outcome of different action options. Experiments show ViMo's ability to generate visually plausible and functionally effective GUIs that enable App agents to make more informed decisions.

Flow-based vision-language-action (VLA) models excel in embodied control but suffer from intractable likelihoods during multi-step sampling, hindering online reinforcement learning. We propose -StepNFT (Step-wise Negative-aware Fine-Tuning), a critic-and-likelihood-free framework that requires only a single forward pass per optimization step and eliminates auxiliary value networks. We identify that wider exploration spaces necessitate finer-grained, step-wise guidance for alignment. Empirically, π-StepNFT unlocks latent potential on LIBERO with competitive few-shot robustness. Moreover, it achieves superior generalization on ManiSkill, outperforming value-based baselines in OOD scenarios by preventing overfitting to multimodal features. This property offers a scalable solution promising for complex real-world applications.