Towards a Science of Collective AI: LLM-based Multi-Agent Systems Need a Transition from Blind Trial-and-Error to Rigorous Science

Recent advancements in Large Language Models (LLMs) have greatly extended the capabilities of Multi-Agent Systems (MAS), demonstrating significant effectiveness across a wide range of complex and open-ended domains. However, despite this rapid progress, the field still relies heavily on empirical trial-and-error. It lacks a unified and principled scientific framework necessary for systematic optimization and improvement. This bottleneck stems from the ambiguity of attribution: first, the absence of a structured taxonomy of factors leaves researchers restricted to unguided adjustments; second, the lack of a unified metric fails to distinguish genuine collaboration gain from mere resource accumulation. In this paper, we advocate for a transition to design science through an integrated framework. We advocate to establish the collaboration gain metric ($Γ$) as the scientific standard to isolate intrinsic gains from increased budgets. Leveraging $Γ$, we propose a factor attribution paradigm to systematically identify collaboration-driving factors. To support this, we construct a systematic MAS factor library, structuring the design space into control-level presets and information-level dynamics. Ultimately, this framework facilitates the transition from blind experimentation to rigorous science, paving the way towards a true science of Collective AI.

UniReason 1.0: A Unified Reasoning Framework for World Knowledge Aligned Image Generation and Editing

Unified multimodal models often struggle with complex synthesis tasks that demand deep reasoning, and typically treat text-to-image generation and image editing as isolated capabilities rather than interconnected reasoning steps. To address this, we propose UniReason, a unified framework that harmonizes these two tasks through a dual reasoning paradigm. We formulate generation as world knowledge-enhanced planning to inject implicit constraints, and leverage editing capabilities for fine-grained visual refinement to further correct visual errors via self-reflection. This approach unifies generation and editing within a shared representation, mirroring the human cognitive process of planning followed by refinement. We support this framework by systematically constructing a large-scale reasoning-centric dataset (~300k samples) covering five major knowledge domains (e.g., cultural commonsense, physics, etc.) for planning, alongside an agent-generated corpus for visual self-correction. Extensive experiments demonstrate that UniReason achieves advanced performance on reasoning-intensive benchmarks such as WISE, KrisBench and UniREditBench, while maintaining superior general synthesis capabilities.

CURP: Codebook-based Continuous User Representation for Personalized Generation with LLMs

User modeling characterizes individuals through their preferences and behavioral patterns to enable personalized simulation and generation with Large Language Models (LLMs) in contemporary approaches. However, existing methods, whether prompt-based or training-based methods, face challenges in balancing personalization quality against computational and data efficiency. We propose a novel framework CURP, which employs a bidirectional user encoder and a discrete prototype codebook to extract multi-dimensional user traits. This design enables plug-and-play personalization with a small number of trainable parameters (about 20M parameters, about 0.2\% of the total model size). Through extensive experiments on variant generation tasks, we show that CURP achieves superior performance and generalization compared to strong baselines, while offering better interpretability and scalability. The code are available at https://github.com/RaidonWong/CURP_code

Strong Reasoning Isn't Enough: Evaluating Evidence Elicitation in Interactive Diagnosis

Interactive medical consultation requires an agent to proactively elicit missing clinical evidence under uncertainty. Yet existing evaluations largely remain static or outcome-centric, neglecting the evidence-gathering process. In this work, we propose an interactive evaluation framework that explicitly models the consultation process using a simulated patient and a \rev{simulated reporter} grounded in atomic evidences. Based on this representation, we introduce Information Coverage Rate (ICR) to quantify how completely an agent uncovers necessary evidence during interaction. To support systematic study, we build EviMed, an evidence-based benchmark spanning diverse conditions from common complaints to rare diseases, and evaluate 10 models with varying reasoning abilities. We find that strong diagnostic reasoning does not guarantee effective information collection, and this insufficiency acts as a primary bottleneck limiting performance in interactive settings. To address this, we propose REFINE, a strategy that leverages diagnostic verification to guide the agent in proactively resolving uncertainties. Extensive experiments demonstrate that REFINE consistently outperforms baselines across diverse datasets and facilitates effective model collaboration, enabling smaller agents to achieve superior performance under strong reasoning supervision. Our code can be found at https://github.com/NanshineLoong/EID-Benchmark .

MAGNET: Towards Adaptive GUI Agents with Memory-Driven Knowledge Evolution

Mobile GUI agents powered by large foundation models enable autonomous task execution, but frequent updates altering UI appearance and reorganizing workflows cause agents trained on historical data to fail. Despite surface changes, functional semantics and task intents remain fundamentally stable. Building on this insight, we introduce MAGNET, a memory-driven adaptive agent framework with dual-level memory: stationary memory linking diverse visual features to stable functional semantics for robust action grounding and procedural memory capturing stable task intents across varying workflows. We propose a dynamic memory evolution mechanism that continuously refines both memories by prioritizing frequently accessed knowledge. Online benchmark AndroidWorld evaluations show substantial improvements over baselines, while offline benchmarks confirm consistent gains under distribution shifts. These results validate that leveraging stable structures across interface changes improves agent performance and generalization in evolving software environments.

CommunityBench: Benchmarking Community-Level Alignment across Diverse Groups and Tasks

Large language models (LLMs) alignment ensures model behaviors reflect human value. Existing alignment strategies primarily follow two paths: one assumes a universal value set for a unified goal (i.e., one-size-fits-all), while the other treats every individual as unique to customize models (i.e., individual-level). However, assuming a monolithic value space marginalizes minority norms, while tailoring individual models is prohibitively expensive. Recognizing that human society is organized into social clusters with high intra-group value alignment, we propose community-level alignment as a "middle ground". Practically, we introduce CommunityBench, the first large-scale benchmark for community-level alignment evaluation, featuring four tasks grounded in Common Identity and Common Bond theory. With CommunityBench, we conduct a comprehensive evaluation of various foundation models on CommunityBench, revealing that current LLMs exhibit limited capacity to model community-specific preferences. Furthermore, we investigate the potential of community-level alignment in facilitating individual modeling, providing a promising direction for scalable and pluralistic alignment.

PersonaDual: Balancing Personalization and Objectivity via Adaptive Reasoning

As users increasingly expect LLMs to align with their preferences, personalized information becomes valuable. However, personalized information can be a double-edged sword: it can improve interaction but may compromise objectivity and factual correctness, especially when it is misaligned with the question. To alleviate this problem, we propose PersonaDual, a framework that supports both general-purpose objective reasoning and personalized reasoning in a single model, and adaptively switches modes based on context. PersonaDual is first trained with SFT to learn two reasoning patterns, and then further optimized via reinforcement learning with our proposed DualGRPO to improve mode selection. Experiments on objective and personalized benchmarks show that PersonaDual preserves the benefits of personalization while reducing interference, achieving near interference-free performance and better leveraging helpful personalized signals to improve objective problem-solving.

SpatialNav: Leveraging Spatial Scene Graphs for Zero-Shot Vision-and-Language Navigation

Although learning-based vision-and-language navigation (VLN) agents can learn spatial knowledge implicitly from large-scale training data, zero-shot VLN agents lack this process, relying primarily on local observations for navigation, which leads to inefficient exploration and a significant performance gap. To deal with the problem, we consider a zero-shot VLN setting that agents are allowed to fully explore the environment before task execution. Then, we construct the Spatial Scene Graph (SSG) to explicitly capture global spatial structure and semantics in the explored environment. Based on the SSG, we introduce SpatialNav, a zero-shot VLN agent that integrates an agent-centric spatial map, a compass-aligned visual representation, and a remote object localization strategy for efficient navigation. Comprehensive experiments in both discrete and continuous environments demonstrate that SpatialNav significantly outperforms existing zero-shot agents and clearly narrows the gap with state-of-the-art learning-based methods. Such results highlight the importance of global spatial representations for generalizable navigation.

SpeechMedAssist: Efficiently and Effectively Adapting Speech Language Models for Medical Consultation

Medical consultations are intrinsically speech-centric. However, most prior works focus on long-text-based interactions, which are cumbersome and patient-unfriendly. Recent advances in speech language models (SpeechLMs) have enabled more natural speech-based interaction, yet the scarcity of medical speech data and the inefficiency of directly fine-tuning on speech data jointly hinder the adoption of SpeechLMs in medical consultation. In this paper, we propose SpeechMedAssist, a SpeechLM natively capable of conducting speech-based multi-turn interactions with patients. By exploiting the architectural properties of SpeechLMs, we decouple the conventional one-stage training into a two-stage paradigm consisting of (1) Knowledge & Capability Injection via Text and (2) Modality Re-alignment with Limited Speech Data, thereby reducing the requirement for medical speech data to only 10k synthesized samples. To evaluate SpeechLMs for medical consultation scenarios, we design a benchmark comprising both single-turn question answering and multi-turn simulated interactions. Experimental results show that our model outperforms all baselines in both effectiveness and robustness in most evaluation settings.

HardcoreLogic: Challenging Large Reasoning Models with Long-tail Logic Puzzle Games

Large Reasoning Models (LRMs) have demonstrated impressive performance on complex tasks, including logical puzzle games that require deriving solutions satisfying all constraints. However, whether they can flexibly apply appropriate rules to varying conditions, particularly when faced with non-canonical game variants, remains an open question. Existing corpora focus on popular puzzles like 9x9 Sudoku, risking overfitting to canonical formats and memorization of solution patterns, which can mask deficiencies in understanding novel rules or adapting strategies to new variants. To address this, we introduce HardcoreLogic, a challenging benchmark of over 5,000 puzzles across 10 games, designed to test the robustness of LRMs on the "long-tail" of logical games. HardcoreLogic systematically transforms canonical puzzles through three dimensions: Increased Complexity (IC), Uncommon Elements (UE), and Unsolvable Puzzles (UP), reducing reliance on shortcut memorization. Evaluations on a diverse set of LRMs reveal significant performance drops, even for models achieving top scores on existing benchmarks, indicating heavy reliance on memorized stereotypes. While increased complexity is the dominant source of difficulty, models also struggle with subtle rule variations that do not necessarily increase puzzle difficulty. Our systematic error analysis on solvable and unsolvable puzzles further highlights gaps in genuine reasoning. Overall, HardcoreLogic exposes the limitations of current LRMs and establishes a benchmark for advancing high-level logical reasoning.