Persistent AI assistants, such as OpenClaw, accumulate large collections of related memories over long-term interactions. As these memories grow, they may reinforce one another, diverge across contexts, or directly conflict, making correct assistance depend on memory relations rather than isolated recall. Existing long-term memory benchmarks do not systematically probe how agents preserve and utilize such relations during downstream tasks. To address this gap, we introduce SubtleMemory, a benchmark for fine-grained relational memory discrimination in long-running AI agents. SubtleMemory constructs relation-controlled latent semantic artifacts whose variants instantiate complementary, nuanced, or contradictory relations, and embeds them into realistic user-agent histories, requiring agents to recover distributed relational structures during later queries and instructions.The benchmark contains 1,522 evaluation instances over 10 long histories, grounded in 1,090 relation-controlled memory-variant sets and spanning user-related and non-user-related queries. Evaluating six standalone memory systems, two Claw-style agents with native memory modules, and three Claw-style agents with plugin memory modules, we find that current systems remain weak on fine-grained relational memory discrimination. We further introduce diagnostic protocols that reveal distinct capability profiles across memory preservation, retrieval, and downstream reasoning stages.

Test-time tree search improves the reasoning capability of large language models under fixed inference budgets. However, existing solution-level MCTS often suffers from local homogenization: once a promising solution pattern is found, subsequent search tends to refine nearby variants rather than explore distinct reasoning paths. This wastes computation on redundant candidates and limits solution-space coverage. To address this problem, we propose Diversity-Aware Reasoning Tree Search (DARTS), a Bayesian tree-search framework that treats reasoning-path diversity as an explicit search objective. DARTS separates expansion into local refinement and orthogonal regeneration, and updates action posteriors using a joint reward over task quality and reasoning-path diversity. On LiveCodeBench and CodeContests, DARTS achieves the best exploration performance and produces more diverse correct solutions. With half the budget, it matches or exceeds the strongest tree-search baseline in four of six settings. On molecular optimization tasks, DARTS improves both optimization quality and structural diversity, suggesting cross-domain transfer.

Planning is central to LLM agents: before acting, an agent must decompose goals, select tools, reason over constraints, and decide when a task is infeasible. Yet existing agent evaluations often report only end-to-end success, making it difficult to determine whether failures stem from planning or execution. We introduce Agent Planning Benchmark (APB), a planning-specific diagnostic benchmark with 4,209 multimodal cases across 22 domains and five settings, covering holistic planning, feedback-conditioned step-wise planning, and robustness under extraneous tools, broken tools, and unsolvable tasks. Across 12 MLLMs, APB reveals systematic weaknesses in long-horizon planning, tool-noise robustness, calibrated refusal, and inference-time refinement. We further validate APB on 200 ToolSandbox tasks and 200 𝜏²-bench tasks, where APB-guided refinement consistently improves plan correctness, plan grade, and downstream execution metrics across three representative models. APB thus serves as an upstream diagnostic complement to execution benchmarks. The APB benchmark and code are available in this URL.

With the advancement of large language models (LLMs) and agent technologies, social bots have shed traditional stereotypes and evolved into autonomous, highly intelligent social media agents. This paper studies whether current social bot detection research is sufficient to identify such agents and discusses potential detection directions. We construct a social media agent dataset from authentic agent accounts on Weibo and Moltbook to evaluate prior detection approaches. Experimental results across various detection algorithms reveal that existing methods suffer from out-of-distribution (OOD) failures and exhibit a cognitive blind spot regarding agent-specific features. Through multidimensional feature comparisons, we find that such agents counterintuitively do not represent a transitional form between humans and traditional bots, but instead follow a distinct pattern that explains the observed generalization failures and feature limitations. Finally, motivated by the excellent performance of LLMs, we propose Agent-Detect-Agent, an online detection agent with tool calling. Experiments demonstrate that this agent further enhances the detection capability of base LLMs and achieves consistently good performance across different account types.

Failures are inevitable when embodied agents execute complex tasks. Visual-language models (VLMs) serve as the core component of embodied agents in perceiving the environment and making decisions. Assessing the capabilities of VLMs in detecting and reasoning about failures has become increasingly important. Previous work primarily considered low-level manipulation failures (e.g., 3cm grasp offsets), neglecting high-level failures arising during long-horizon task execution (e.g., object-dropping failure in the “clean room” task) by embodied agents. In this paper, we propose FAER, a failure-aware benchmark aiming to evaluate the performance of VLMs in terms of failure detection, failure categorization, failure description, and failure correction in long-horizon tasks. FAER comprises 3,323 episodes, spanning 3 scenes, 65 tasks, and 83 objects. We assess the performance of 16 widely utilized VLMs and 4 LLMs for FAER tasks. Experimental results show that nearly all VLMs, even GPT-4o, exhibit limited performance in failure detection with a high false negative rate, meaning that they tend to ignore abnormal events, revealing notable gaps in current models’ capacity to effectively handle failures.

Embodied planning is dedicated to the goal of building agents capable of perceiving, reasoning, and executing long-horizon tasks in physical environments. However, existing embodied benchmarks predominantly focus on single-agent execution or restricted multi-agent settings, leaving long-horizon tasks that involve complex inter-agent dependencies largely underexplored despite their prevalence in real-world embodied scenarios. To address this gap, we introduce CoEmbod, the first realistic 3D simulated embodied benchmark for long-horizon multi-agent embodied cooperation with diverse inter-agent dependencies. CoEmbod contains 155 expert-curated tasks across 40 scenes and 10845 objects, covering three cooperation regimes: spatial distribution, parallel constraint, and failure recovery. These regimes cover a broad spectrum of cooperative requirements, ranging from distributed task allocation to synchronized execution and assisted failure recovery. CoEmbod further provides a process-level evaluation framework with Budgeted Cooperative Success (BCS), which jointly measures goal completion, cooperation fidelity, and execution efficiency. Experiments reveal that even the leading LMM powering a well-established agentic workflow achieves only 20.23% success, highlighting the difficulty of long-horizon multi-agent cooperation with diverse inter-agent dependencies in complex embodied environments.

Social bot accounts have long been disseminating disinformation and engaging in malicious activities on social media platforms. Detecting these social bots has become a critical and urgent task, essential for maintaining a healthy online ecosystem. Existing social bot detection research usually provides detection results directly without corresponding supportive explanations, making it difficult to assess the extent to which such predictions are trustworthy. This is a key concern for online moderation. In this work, we explore the detection interpretation and summarize a four-dimensional clue framework from individual and social perspectives. We propose CDRBot, which primarily employs outcome-reward reinforcement learning to train inspectors to generate faithful, grounded, and readable clues from the User Information, Semantic Features, Interactive Situation, and Behavioral Pattern. These clues are then integrated to make final predictions. Experimental results demonstrate that our approach outperforms other baselines in detection performance. The generated clues are faithful, grounded, and readable, and can significantly enhance the performance of large language models in social bot detection.

Large language models (LLMs) often succumb to users’ viewpoints when faced with conflicting perspectives. We identify two key biases underlying this issue : stance homogeneity bias and human preference bias. To address these biases, we propose a novel two-stage training framework: Multi-stance Discussion Sampling and Truth Alignment Training (MDTA). First, we introduce an equal multi-stance discussion framework to automatically generate multi-model discussion datasets. Based on this framework, we construct the first and largest multi-model fair discussion dataset named Eq-Discussion for supervised fine-tuning, reducing stance homogeneity bias. Second, we optimize Reinforcement Learning from Human Feedback (RLHF) to align with discussion correctness, mitigating human preference bias. Extensive experimental results demonstrate that MDTA effectively reduces both biases and significantly enhances the performance of LLMs across a variety of downstream tasks, including reading comprehension, logical reasoning, and social question answering. Furthermore, we observe that MDTA improves the generalization capabilities of LLMs, leading to substantial performance improvements in non-discussion scenarios and on out-of-domain datasets.

With the widespread deployment of generative language models, concerns about safety issues have continuously grown. High-quality fine-tuning data generated from red teaming plays a crucial role in the model’s safety. Recently, automated red teaming approaches have been proposed to create test cases. However, these approaches, which rely on open-ended generation, encounter issues related to inefficiency and low attack success rates. In this work, we introduce a black-box approach that ingeniously exploits the unique properties of the nullspace to disentangle and regulate the crucial success information within test cases. Our study provides a brand-new perspective for automated red team research. Experimental results demonstrate that our approach outperforms baseline methods regarding the attack success rate. The generated test cases also excel in aspects of diversity and fluency.

Although Retrieval-Augmented Large Language Models (RALMs) demonstrate their superiority in terms of factuality, they do not consistently outperform the original retrieval-free Language Models (LMs). Our experiments reveal that this example-level performance inconsistency exists not only between retrieval-augmented and retrieval-free LM but also among different retrievers. To understand this phenomenon, we investigate the degeneration behavior of RALMs and theoretically decompose it into four categories. Further analysis based on our decomposition reveals that the innate difference in knowledge sources and the unpredictable degeneration of the reader model contribute most to the inconsistency. Drawing from our analysis, we introduce Ensemble of Retrievers (EoR), a trainable framework that can adaptively retrieve from different knowledge sources and effectively decrease unpredictable reader errors. Our experiments on Open Domain Question Answering show that EoR substantially improves performance over the RALM with a single retriever by considerably reducing inconsistent behaviors.

Argumentative essay generation (AEG) aims to generate complete texts on specific controversial topics or debates. Although current AEG methods can generate individual opinions, they often overlook the high-level connections between these opinions. This often leads to the generated results being mired in logical confusion, unable to proof their own arguments effectively. The generated essay may present evidence that contradicts the claims or they may fail to assemble the claims into logical flow. In this paper, we present a unified two-stage framework: Proof-Enhancement and Self-Annotation (PESA) for AEG with a focus on logical enhancement. Specifically, we first construct pseudo-labels for logical information,claims and grounds, using a large language model. We then propose a tree planning approach that introduces proof principles and ensures logical consistency. Extensive experimental results show that, benefiting from proof principle guidance, PESA generates argumentative essays with better logical validity and persuasiveness than strong baseline models.

CLIP has shown a remarkable zero-shot capability on a wide range of vision tasks. Previously, CLIP is only regarded as a powerful visual encoder. However, after being pre-trained by language supervision from a large amount of image-caption pairs, CLIP itself should also have acquired some few-shot abilities for vision-language tasks. In this work, we empirically show that CLIP can be a strong vision-language few-shot learner by leveraging the power of language. We first evaluate CLIP’s zero-shot performance on a typical visual question answering task and demonstrate a zero-shot cross-modality transfer capability of CLIP on the visual entailment task. Then we propose a parameter-efficient fine-tuning strategy to boost the few-shot performance on the vqa task. We achieve competitive zero/few-shot results on the visual question answering and visual entailment tasks without introducing any additional pre-training procedure.