While autonomous software engineering (SWE) agents are reshaping programming paradigms, they currently suffer from a “closed-world” limitation: they attempt to fix bugs from scratch or solely using information from the current repository, ignoring the immense historical human experience available on platforms like GitHub. Accessing this open-world experience is hindered by the unstructured and fragmented nature of real-world issue-tracking data. In this paper, we introduce MemGovern, a framework designed to govern and transform raw GitHub data into actionable experiential memory for agents. MemGovern employs experience governance to convert human experience into agent-friendly experience cards and introduces an agentic experience search strategy that enables logic-driven retrieval of human expertise. By producing 135K governed experience cards, MemGovern achieves a significant performance boost, improving resolution rates on the SWE-bench benchmark by 4.65%. As a plug-in approach, MemGovern provides a solution for agent-friendly memory infrastructure.

Self-evolution methods enhance code generation through iterative ”generate-verify-refine” cycles, yet existing approaches suffer from low exploration efficiency, failing to discover algorithmically superior solutions within limited budgets. We identify the root cause as the lack of systematic controllability: execution feedback is not reliably converted into targeted search decisions across route planning, edit granularity, and experience reuse. We propose Controlled Self-Evolution (CSE), a framework that couples three components into an integrated control architecture. Diversified Planning Initialization provides route diversity through structurally distinct algorithmic strategies. Modular Evolution provides feedback attribution and module-level editing by decomposing programs into functional modules and applying targeted mutation and compositional crossover. Hierarchical Evolution Memory maintains search state within tasks and distills reusable optimization patterns across tasks. Experiments on EffiBench-X demonstrate that CSE consistently outperforms all baselines across various LLM backbones. Further analysis reveals non-additive synergy among control layers, confirming that CSE’s effectiveness arises from component coupling rather than independent contributions. The code and evaluation implementation are publicly available at https://github.com/QuantaAlpha/EvoControl.

As Large Language Models (LLMs) evolve from static dialogue interfaces to autonomous general agents, effective memory is paramount to ensuring long-term consistency. However, existing benchmarks primarily focus on casual conversation or task-oriented dialogue, failing to capture “long-term project-oriented” interactions where agents must track evolving goals. To bridge this gap, we introduce RealMem, the first benchmark grounded in realistic project scenarios. RealMem comprises over 2,000 cross-session dialogues across eleven scenarios, utilizing natural user queries for evaluation. We propose a synthesis pipeline that integrates Project Foundation Construction, Multi-Agent Dialogue Generation, and Memory and Schedule Management to simulate the dynamic evolution of memory. Experiments reveal that current memory systems face significant challenges in managing the long-term project states and dynamic context dependencies inherent in real-world projects. Our code and datasets are available at https://anonymous.4open.science/r/realmem-A1E4.

AI coding has emerged as a core application of large language models (LLMs), evolving from single-file coding tasks towards complex software engineering (SWE) scenarios. Recent advances in agents have enabled multi-file, multi-language, and dependency-aware AI coding, significantly expanding the scope of AI-assisted software development. While a variety of benchmarks have been proposed to evaluate coding capabilities in general-purpose or GPU coding ecosystems such as CUDA and ROCm, systematic evaluation for Huawei Ascend CANN remains largely underexplored. In this work, we propose LiveCANNBench, an SWE-level benchmark designed for AI coding in the CANN software stack. LiveCANNBench is constructed from real-world CANN repositories and consists of over 400 task instances spanning multi-file, multi-language, and execution-aware coding challenges. Unlike existing static benchmarks that primarily focus on kernel-level code generation, LiveCANNBench adopts a live benchmarking paradigm, effectively mitigating data leakage and enabling more reliable evaluation of modern coding agents.

Questions about a recurring person in long videos are rarely answered from a single moment. They require evidence that is sufficient for the claim, comparisons across episodes, synthesis over recurring audiovisual patterns, and refusal when the source material does not support an answer. Existing long-video benchmarks do not isolate these requirements. We introduce KnowMeBenchV2, an evaluation-only benchmark over five public multi-episode video series, with 327 videos and 3,061 questions covering four evidence scopes: single-clip extraction, temporal and event relations, cross-episode ordering, and cross-episode persona synthesis. A separate grounded-abstention task serves as a contrastive support check: its questions are surface-similar to supported person-linked questions but are not licensed by the source evidence. The most interpretive synthesis items are human-authored and quality-audited by trained annotators with full source access. A matched human reference reaches 78.0 on cross-episode synthesis. To diagnose failures, we compare evidence routes under a fixed Qwen3.5-9B answerer and a task-matched pre-tokenization character routing budget. ASR-Retrieval reaches 2.7 on synthesis, while KnowMe reaches 64.9. In this diagnostic, the final answerer reasons over textualized evidence constructed from audiovisual sources rather than raw video tokens. The gap therefore isolates the importance of audiovisual evidence construction and cross-episode organization before answer generation.

Visual scale recognition is a fundamental aspect for humans to perceive physical quantities in the real world, and it is crucial for enabling human-like intelligence in multimodal large language models (MLLMs). However, existing benchmarks typically focus on a single type of quantity (e.g., time) or a specific format (e.g., dials), lacking a comprehensive evaluation of scale recognition capabilities. To address these problems, we propose ScaleBench, a visual scale recognition benchmark built using images from COCO, Open Images, and Flickr, designed to comprehensively evaluate the scale recognition capabilities of MLLMs. To ensure high data quality, we develop detailed annotation guidelines and procedures, resulting in a total of 6,574 annotated samples. Based on this benchmark, we evaluate multiple closed-source and open-source MLLMs. Experimental results reveal that the best-performing model achieves only 42.60% accuracy, far lower than the 97.40% of humans. Furthermore, we conduct in-depth experimental analyses and provide future research directions. Our benchmark and implementation codes are available at https://github.com/Sonder-hang/ScaleBench.

Multimodal large language models (MLLMs) have achieved remarkable progress in recent years, yet their ability to perform left–right reasoning in mirror contexts—a fundamental element of spatial cognition—remains underexplored. To address this gap, we introduce MirrorQA, a manually constructed benchmark with 5,549 samples, designed to evaluate MLLMs’ capability to distinguish left from right from a subject-centered perspective. MirrorQA is built through a three-stage pipeline (annotation, verification, and final review) to ensure high-quality labeling. Comprehensive evaluations on both open- and closed-source MLLMs show that even the best-performing models achieve only 65.40% accuracy, far below the 99.28% accuracy of humans. These results highlight substantial challenges in current MLLMs when reasoning about left and right, and point to promising directions for future research. MirrorQA and its code are publicly available at anonymous link https://github.com/stargazer-zeno/MirrorQA.

AI Clones aim to simulate an individual’s thoughts and behaviors to enable long-term, personalized interaction, placing stringent demands on memory systems to model experiences, emotions, and opinions over time. Existing memory benchmarks primarily rely on user–agent conversational histories, which are temporally fragmented and insufficient for capturing continuous life trajectories. We introduce CloneMem, a benchmark for evaluating long-term memory in AI Clone scenarios grounded in non-conversational digital traces, including diaries, social media posts, and emails, spanning one to three years. CloneMem adopts a top-down data construction framework to ensure longitudinal coherence and defines tasks that assess an agent’s ability to track evolving personal states. Experiments show that current memory mechanisms struggle in this setting, highlighting open challenges for life-grounded personalized AI. Code and dataset are available at https://github.com/AvatarMemory/CloneMemBench

Existing long-horizon memory benchmarks mostly use multi-turn dialogues or synthetic user histories, which makes retrieval performance an imperfect proxy for person understanding. We present Knowme-Bench, a publicly releasable benchmark built from long-form autobiographical narratives, where actions, context, and inner thoughts provide dense evidence for inferring stable motivations and decision principles. Knowme-Bench reconstructs each narrative into a flashback-aware, time-anchored stream and evaluates models with evidence-linked questions spanning factual recall, subjective state attribution, and principle-level reasoning. Across diverse narrative sources, retrieval-augmented systems mainly improve factual accuracy, while errors persist on temporally grounded explanations and higher-level inferences, highlighting the need for memory mechanisms beyond retrieval.

Current defenses for Large Language Models (LLMs) often suffer from a ”memory gap”: parameter-modifying methods are computationally rigid, while inference-time filters cannot retain or reuse defense knowledge across interactions. To address this, we propose SafetyMem, a novel framework that secures LLMs through a dual-component safety memory system. SafetyMem consists of Semantic Safety Memory (SSM), which consolidates diverse jailbreak attempts into a structured knowledge base of attack patterns, and Episodic Safety Memory (ESM), which maintains an evolving set of procedural rules refined from historical detection failures. Unlike static defenses, SafetyMem allows the model to ”remember” and adapt to emerging adversarial strategies without parameter retraining. To further enhance robustness, we introduce an adversarial memory expansion mechanism that proactively generates challenging variants to solidify these memories. Experiments on standard and stealthy jailbreak benchmarks show that SafetyMem substantially reduces attack success rates while preserving efficiency and interpretability, consistently outperforming state-of-the-art baselines across multiple LLMs.

Large language models (LLMs) face significant challenges in sustaining long-term memory for agentic applications due to limited context windows. To address this limitation, many work has proposed diverse memory mechanisms to support long-term, multi-turn interactions, leveraging different approaches tailored to distinct memory storage objects, such as KV caches. In this survey, we present a unified taxonomy that organizes memory systems for long-context scenarios by decoupling memory abstractions from model-specific inference and training methods. We categorize LLM memory into three primary paradigms: natural language tokens, intermediate representations and parameters. For each paradigm, we organize existing methods by three management stages, including memory construction, update, and query, so that long-context memory mechanisms can be described in a consistent way across system designs, with their implementation choices and constraints made explicit. Finally, we outline key research directions for long-context memory system design.

Recent diffusion and autoregressive video generators have substantially improved short-form synthesis, yet composing multiple independently generated clips into a coherent audiovisual story remains challenging. Character identities may drift across scenes, narrative events can become inconsistent, and dialogue or sound may not align with the evolving visual context. We present **DramaAgent**, a hierarchical, agentic, and model-agnostic framework for multi-scene long-horizon audiovisual story generation. Rather than modifying the underlying video backbone, DramaAgent introduces a stateful control layer that combines hierarchical story planning, persistent character conditioning, scene-wise generation, and reflection-guided targeted repair. The framework maintains reusable story and character states across scenes, diagnoses failures along identity, semantic, temporal, and audio-visual dimensions, and maps each diagnosed failure to a corresponding localized repair operation. Experiments on 50 narrative scripts across five video-generation backbones show consistent improvements in semantic alignment, character consistency, narrative fidelity, and scene-level audio-visual compatibility. Controlled same-backbone comparisons further show that these gains cannot be explained solely by stronger generators or additional sampling. Our results suggest that persistent state, failure diagnosis, and targeted repair provide a practical control paradigm for coherent multi-scene audiovisual generation.