Weekly Digest — 2026-W15

Visual trace replay for AI apps to fix bugs in one click

Your best QA team — 9x faster, 20х cheaper

Everything in OpenClaw's terminal, you can now do visually

The AI video model that actually feels alive.

Turn support emails into tickets

Track the Artemis II mission from your Mac

Automate what APIs can't in one prompt done locally

Track body measurements, fat%, lean Mass, progress & more

A full navigation utility in ~2MB

The intelligence layer for ChatGPT Ads

The AI co-author for research papers

Select any screen area, get text instantly

On-policy distillation (OPD) has become a popular training paradigm in the LLM community. This paradigm selects a larger model as the teacher to provide dense, fine-grained signals for each sampled trajectory, in contrast to reinforcement learning with verifiable rewards (RLVR), which only obtains sparse signals from verifiable outcomes in the environment. Recently, the community has explored on-policy self-distillation (OPSD), where the same model serves as both teacher and student, with the teacher receiving additional privileged information such as reference answers to enable self-evolution. This paper demonstrates that learning signals solely derived from the privileged teacher result in severe information leakage and unstable long-term training. Accordingly, we identify the optimal niche for self-distillation and propose RLSD (RLVR with Self-Distillation). Specifically, we leverage self-distillation to obtain token-level policy differences for determining fine-grained update magnitudes, while continuing to use RLVR to derive reliable update directions from environmental feedback (e.g., response correctness). This enables RLSD to simultaneously harness the strengths of both RLVR and OPSD, achieving a higher convergence ceiling and superior training stability.

Recent streaming video understanding methods increasingly rely on complex memory mechanisms to handle long video streams. We challenge this trend with a simple finding: a sliding-window baseline that feeds only the most recent N frames to an off-the-shelf VLM already matches or surpasses published streaming models. We formalize this baseline as SimpleStream and evaluate it against 13 major offline and online video LLM baselines on OVO-Bench and StreamingBench. Despite its simplicity, SimpleStream delivers consistently strong performance. With only 4 recent frames, it reaches 67.7% average accuracy on OVO-Bench and 80.59% on StreamingBench. Controlled ablations further show that the value of longer context is backbone-dependent rather than uniformly increasing with model scale, and reveal a consistent perception-memory trade-off: adding more historical context can improve recall, but often weakens real-time perception. This suggests that stronger memory, retrieval, or compression modules should not be taken as evidence of progress unless they clearly outperform SimpleStream under the same protocol. We therefore argue that future streaming benchmarks should separate recent-scene perception from long-range memory, so that performance improvements from added complexity can be evaluated more clearly.

Can warping tokens, rather than pixels, help multimodal large language models (MLLMs) understand how a scene appears from a nearby viewpoint? While MLLMs perform well on visual reasoning, they remain fragile to viewpoint changes, as pixel-wise warping is highly sensitive to small depth errors and often introduces geometric distortions. Drawing on theories of mental imagery that posit part-level structural representations as the basis for human perspective transformation, we examine whether image tokens in ViT-based MLLMs serve as an effective substrate for viewpoint changes. We compare forward and backward warping, finding that backward token warping, which defines a dense grid on the target view and retrieves a corresponding source-view token for each grid point, achieves greater stability and better preserves semantic coherence under viewpoint shifts. Experiments on our proposed ViewBench benchmark demonstrate that token-level warping enables MLLMs to reason reliably from nearby viewpoints, consistently outperforming all baselines including pixel-wise warping approaches, spatially fine-tuned MLLMs, and a generative warping method.

Multimodal Large Language Models (MLLMs) are evolving from passive observers into active agents, solving problems through Visual Expansion (invoking visual tools) and Knowledge Expansion (open-web search). However, existing evaluations fall short: they lack flexible tool integration, test visual and search tools separately, and evaluate primarily by final answers. Consequently, they cannot verify if tools were actually invoked, applied correctly, or used efficiently. To address this, we introduce Agentic-MME, a process-verified benchmark for Multimodal Agentic Capabilities. It contains 418 real-world tasks across 6 domains and 3 difficulty levels to evaluate capability synergy, featuring over 2,000 stepwise checkpoints that average 10+ person-hours of manual annotation per task. Each task includes a unified evaluation framework supporting sandboxed code and APIs, alongside a human reference trajectory annotated with stepwise checkpoints along dual-axis: S-axis and V-axis. To enable true process-level verification, we audit fine-grained intermediate states rather than just final answers, and quantify efficiency via an overthinking metric relative to human trajectories. Experimental results show the best model, Gemini3-pro, achieves 56.3% overall accuracy, which falls significantly to 23.0% on Level-3 tasks, underscoring the difficulty of real-world multimodal agentic problem solving.

Modern LLMs scale at test-time, e.g. via repeated sampling, where inference cost grows with model size and the number of samples. This creates a trade-off that pretraining scaling laws, such as Chinchilla, do not address. We present Train-to-Test (T^2) scaling laws that jointly optimize model size, training tokens, and number of inference samples under fixed end-to-end budgets. T^2 modernizes pretraining scaling laws with pass@k modeling used for test-time scaling, then jointly optimizes pretraining and test-time decisions. Forecasts from T^2 are robust over distinct modeling approaches: measuring joint scaling effect on the task loss and modeling impact on task accuracy. Across eight downstream tasks, we find that when accounting for inference cost, optimal pretraining decisions shift radically into the overtraining regime, well-outside of the range of standard pretraining scaling suites. We validate our results by pretraining heavily overtrained models in the optimal region that T^2 scaling forecasts, confirming their substantially stronger performance compared to pretraining scaling alone. Finally, as frontier LLMs are post-trained, we show that our findings survive the post-training stage, making T^2 scaling meaningful in modern deployments.

Humans build shared spatial understanding by communicating partial, viewpoint-dependent observations. We ask whether Multimodal Large Language Models (MLLMs) can do the same, aligning distinct egocentric views through dialogue to form a coherent, allocentric mental model of a shared environment. To study this systematically, we introduce COSMIC, a benchmark for Collaborative Spatial Communication. In this setting, two static MLLM agents observe a 3D indoor environment from different viewpoints and exchange natural-language messages to solve spatial queries. COSMIC contains 899 diverse scenes and 1250 question-answer pairs spanning five tasks. We find a consistent capability hierarchy, MLLMs are most reliable at identifying shared anchor objects across views, perform worse on relational reasoning, and largely fail at building globally consistent maps, performing near chance, even for the frontier models. Moreover, we find thinking capability yields consistent gains in anchor grounding, but is insufficient for higher-level spatial communication. To contextualize model behavior, we additionally collect 250 human-human dialogues. Humans achieve 95% aggregate accuracy, leaving significant room for improvement for even the best performing model Gemini-3-Pro-Thinking which achieves 72% aggregate accuracy. Moreover, human conversations become increasingly specific as partners converge on a shared mental model, whereas model dialogues continue to explore new possibilities rather than converging, consistent with a limited ability to build and maintain a robust shared mental model. Our code and data is available at https://github.com/ankursikarwar/Cosmic

World models have garnered significant attention as a promising research direction in artificial intelligence, yet a clear and unified definition remains lacking. In this paper, we introduce OpenWorldLib, a comprehensive and standardized inference framework for Advanced World Models. Drawing on the evolution of world models, we propose a clear definition: a world model is a model or framework centered on perception, equipped with interaction and long-term memory capabilities, for understanding and predicting the complex world. We further systematically categorize the essential capabilities of world models. Based on this definition, OpenWorldLib integrates models across different tasks within a unified framework, enabling efficient reuse and collaborative inference. Finally, we present additional reflections and analyses on potential future directions for world model research. Code link: https://github.com/OpenDCAI/OpenWorldLib

Current document parsing methods compete primarily on model architecture innovation, while systematic engineering of training data remains underexplored. Yet SOTA models of different architectures and parameter scales exhibit highly consistent failure patterns on the same set of hard samples, suggesting that the performance bottleneck stems from shared deficiencies in training data rather than architecture itself. Building on this finding, we present \minerupro, which advances the state of the art solely through data engineering and training strategy optimization while keeping the 1.2B-parameter architecture of \mineru completely fixed. At its core is a Data Engine co-designed around coverage, informativeness, and annotation accuracy: Diversity-and-Difficulty-Aware Sampling expands training data from under 10M to 65.5M samples while correcting distribution shift; Cross-Model Consistency Verification leverages output agreement among heterogeneous models to assess sample difficulty and generate reliable annotations; the Judge-and-Refine pipeline improves annotation quality for hard samples through render-then-verify iterative correction. A three-stage progressive training strategy -- large-scale pre-training, hard sample fine-tuning, and GRPO alignment -- sequentially exploits these data at different quality tiers. On the evaluation front, we fix element-matching biases in OmniDocBench~v1.5 and introduce a Hard subset, establishing the more discriminative OmniDocBench~v1.6 protocol. Without any architectural modification, \minerupro achieves 95.69 on OmniDocBench~v1.6, improving over the same-architecture baseline by 2.71 points and surpassing all existing methods including models with over 200times more parameters.

Vision-Language-Action (VLA) models achieve strong performance in robotic manipulation by leveraging pre-trained vision-language backbones. However, in downstream robotic settings, they are typically fine-tuned with limited data, leading to overfitting to specific instruction formulations and leaving robustness to paraphrased instructions underexplored. To study this gap, we introduce LIBERO-Para, a controlled benchmark that independently varies action expressions and object references for fine-grained analysis of linguistic generalization. Across seven VLA configurations (0.6B-7.5B), we observe consistent performance degradation of 22-52 pp under paraphrasing. This degradation is primarily driven by object-level lexical variation: even simple synonym substitutions cause large drops, indicating reliance on surface-level matching rather than semantic grounding. Moreover, 80-96% of failures arise from planning-level trajectory divergence rather than execution errors, showing that paraphrasing disrupts task identification. Binary success rate treats all paraphrases equally, obscuring whether models perform consistently across difficulty levels or rely on easier cases. To address this, we propose PRIDE, a metric that quantifies paraphrase difficulty using semantic and syntactic factors. Our benchmark and corresponding code are available at: https://github.com/cau-hai-lab/LIBERO-Para

Extended reasoning in large language models (LLMs) creates severe KV cache memory bottlenecks. Leading KV cache compression methods estimate KV importance using attention scores from recent post-RoPE queries. However, queries rotate with position during RoPE, making representative queries very few, leading to poor top-key selection and unstable reasoning. To avoid this issue, we turn to the pre-RoPE space, where we observe that Q and K vectors are highly concentrated around fixed non-zero centers and remain stable across positions -- Q/K concentration. We show that this concentration causes queries to preferentially attend to keys at specific distances (e.g., nearest keys), with the centers determining which distances are preferred via a trigonometric series. Based on this, we propose TriAttention to estimate key importance by leveraging these centers. Via the trigonometric series, we use the distance preference characterized by these centers to score keys according to their positions, and also leverage Q/K norms as an additional signal for importance estimation. On AIME25 with 32K-token generation, TriAttention matches Full Attention reasoning accuracy while achieving 2.5x higher throughput or 10.7x KV memory reduction, whereas leading baselines achieve only about half the accuracy at the same efficiency. TriAttention enables OpenClaw deployment on a single consumer GPU, where long context would otherwise cause out-of-memory with Full Attention.

While textual frequency has been validated as relevant to human cognition in reading speed, its relatedness to Large Language Models (LLMs) is seldom studied. We propose a novel research direction in terms of textual data frequency, which is an understudied topic, to the best of our knowledge. Our framework is composed of three units. First, this paper proposes Textual Frequency Law (TFL), which indicates that frequent textual data should be preferred for LLMs for both prompting and fine-tuning. Since many LLMs are closed-source in their training data, we propose using online resources to estimate the sentence-level frequency. We then utilize an input paraphraser to paraphrase the input into a more frequent textual expression. Next, we propose Textual Frequency Distillation (TFD) by querying LLMs to conduct story completion by further extending the sentences in the datasets, and the resulting corpora are used to adjust the initial estimation. Finally, we propose Curriculum Textual Frequency Training (CTFT) that fine-tunes LLMs in an increasing order of sentence-level frequency. Experiments are conducted on our curated dataset Textual Frequency Paired Dataset (TFPD) on math reasoning, machine translation, commonsense reasoning and agentic tool calling. Results show the effectiveness of our framework.

Video Large Language Models (VideoLLMs) have achieved strong performance on many video understanding tasks, but most existing systems remain offline and are not well-suited for live video streams that require continuous observation and timely response. Recent streaming VideoLLMs have made progress, yet current approaches often rely on decoupled trigger-response pipelines or are limited to captioning-style narration, reducing their effectiveness for open-ended question answering and long-horizon interaction. We propose AURA (Always-On Understanding and Real-Time Assistance), an end-to-end streaming visual interaction framework that enables a unified VideoLLM to continuously process video streams and support both real-time question answering and proactive responses. AURA integrates context management, data construction, training objectives, and deployment optimization for stable long-horizon streaming interaction. It achieves state-of-the-art performance on streaming benchmarks and supports a real-time demo system with ASR and TTS running at 2 FPS on two 80G accelerators. We release the AURA model together with a real-time inference framework to facilitate future research.

New York Times : The rapid adoption of AI coding tools has let workers generate massive volumes of code, leaving companies scrambling to review and secure the AI-generated code —  When a financial services company recently began using Cursor, an artificial intelligence technology that writes computer code, the difference that it made was immediate.

Bloomberg : Sources: OpenAI, Anthropic, and Google are sharing information via the Frontier Model Forum to detect adversarial distillation attempts that violate their ToS —  Rivals OpenAI, Anthropic PBC, and Alphabet Inc.'s Google have begun working together to try to clamp down on Chinese competitors extracting results …

Ian King / Bloomberg : Australian AI infrastructure startup Firmus raised $505M led by Coatue at a $5.5B valuation, bringing its funding raised in the last six months to $1.35B —  Data center builder Firmus Technologies Pty raised $505 million in an investment round led by Coatue Management LLC …

Sarah Nassauer / Wall Street Journal : A look at Eko, whose Arkansas “capture factory” creates digital product catalogs intended to serve as training data for retail-focused AI models —  In an Arkansas ‘capture factory,’ hand models and food stylists are preparing for the future of shopping

Nate Silver / Silver Bulletin : How social media became a freak show: X punishes external links and most top accounts, such as Catturd, are very low-quality but get more engagement than NYT —  The ecosystem is unhealthy, especially on Twitter, and that's producing some strange beasts among the most influential accounts.

Nate Raymond / Reuters : A federal appeals court rules New Jersey cannot block Kalshi users in the state from sports-related event contracts, finding CFTC has exclusive jurisdiction —  A federal appeals court ruled on Monday that New Jersey gaming regulators cannot prevent Kalshi from allowing people in the state …

Carl Franzen / VentureBeat : Z.ai releases GLM-5.1, a 754B-parameter model that it says outperforms GPT-5.4 and Claude Opus 4.6 on SWE-bench Pro, available under an MIT license —  Is China picking back up the open source AI baton?  —  Z.ai, also known as Zhupai AI, a Chinese AI startup best known for its powerful …

Andy Greenberg / Wired : A group of US agencies including the FBI and the NSA warns that Iran-linked hackers have targeted industrial control devices used in US critical infrastructure —  As Trump threatens Iranian infrastructure, the US government warns that Iran has carried out its own digital attacks against US critical infrastructure.

Andrew Romero / 9to5Google : Google rolls out an AI Enhance button for Photos on Android globally, offering automated lighting and contrast adjustments, and video playback speed controls —  Google announced two new additions to Google Photos for all Android users, and they've already begun rolling out.

Jessica Toonkel / Wall Street Journal : Elon Musk amends his OpenAI lawsuit to ask that damages he might win be awarded to OpenAI's charity arm and that Altman be removed from OpenAI's nonprofit board —  Tesla billionaire also seeks Sam Altman's removal from OpenAI nonprofit's board in amendment to suit over for-profit conversion

Michael Nuñez / VentureBeat : Anthropic says Mythos Preview achieves 93.9% on SWE-bench Verified, compared with 80.8% for Opus 4.6, and 77.8% on SWE-bench Pro, versus 53.4% for Opus 4.6 —  Anthropic on Tuesday announced Project Glasswing, a sweeping cybersecurity initiative that pairs an unreleased frontier AI model …

Alex Kantrowitz / Big Technology : Q&A with OpenAI President Greg Brockman about OpenAI's research direction, how far it can push Codex, closing Sora, betting on text vs. world models, and more —  OpenAI is shifting strategies yet again.  Here's the logic behind the latest moves and what they mean for the company's direction.

德国公开了曾在早期运营俄罗斯勒索软件组织 GandCrab 和 REvil 的头目 UNKN 的身份。31 岁的 Daniil Maksimovich Shchukin 于 2019-2021 年间在德国实施了至少 130 起计算机破坏和勒索行动。德国称，Shchukin 以及另一名俄罗斯人——43 岁 Anatoly Sergeevitsch Kravchuk——一起勒索了近 200 万欧元，造成经济损失逾 3500 万欧元。德国联邦刑事警察局（BKA）称他领导的 GandCrab 和 REvil 首创了双重勒索——先向受害者收取赎金提供解锁的密钥，然后再收取一笔费用换取不公开被盗数据的承诺。GandCrab 在 2019 年成功勒索逾 20 亿美元后宣布解散，但随后就以 REvil 的名字再次亮相。

Chrome 和 Firefox 等浏览器都支持延迟加载。延迟加载又名懒加载，顾名思义，为了加快页面加载速度而推迟加载特定对象，这些对象直到要使用时才会开始加载，如 Chrome 从 2019 年起就延迟加载图像和 iframe。现在它正在 Chrome 148 上测试延迟加载视频和音频以改进浏览器性能。今天很多网站尤其是新闻网站都会在页面中嵌入视频和音频，影响了页面加载速度。

今天的司机有很多方法躲避超速相机的监测，比如手机应用程序能提前通知司机前方有测速相机，司机随后放慢车速，通过之后再加速行驶。为了遏制此类行为，越来越多的地方开始推出测量均速的相机系统：跟踪同一辆汽车在多个监控点之间的均速，如果均速超过限速 10 英里/时或以上，则对相关车辆开出罚单。美国科罗拉多州于 2023 年通过法律允许使用自动车辆识别系统计算汽车在不同摄像头之间的均速，去年底交警开始正式对超速行驶的汽车开罚单。罚款为 75 美元，不扣分，罚单将开给车主，不管驾驶汽车的司机是谁。地图软件如 TomTom 也对此采取了应对措施，为司机提供测均速区域的均速信息。

AI 工具的用户通常可分为两类：其一将 AI 视为功能强大但会犯错的服务，需要人类仔细监督和审查以发现其中的推理或事实错误；其二将 AI 视为无所不知——此类用户被称为是“认知投降派”。宾夕法尼亚大学沃顿商学院的研究人员对 1372 名参与者和逾 9500 次测试后发现，高达 73.2% 的情况下参与者愿意接受 AI 错误的推理，只有 19.7% 的情况下会推翻推理。研究人员表示这一结果“表明人很容易将 AI 生成的输出融入到决策过程中，且通常几乎没有任何抵触或怀疑”,“流畅、自信的输出会被视为有认知权威性，从而降低审查门槛，减弱了通常会促使人们进行深思熟虑的元认知信号”。他们发现，倾向于将 AI 视为权威的人更容易被 AI 提供的错误答案误导。

亚马逊 AWS 工程师 Salvatore Dipietro 报告 Linux 7.0 下 PostgreSQL 的吞吐量和延迟性能出现了显著的下降。Linux 7.0 目前还在开发中，预计会在一两周内发布。测试显示，在基于 arm64 架构的 Graviton4 服务器上 PostgreSQL 的吞吐量仅为上个内核版本的 0.51 倍，原因是用户空间自旋锁导致花费的时间大幅增加。根本原因被认为是 Linux 7.0 新引入的对内核可用抢占模式的限制上。PostgreSQL 开发者要求在不同条件下重复进行更多测试。

Canonical 即将于本月晚些时候正式释出的 Ubuntu 26.04 LTS 把最低内存需求提高到了 6GB。Ubuntu 14.04 LTS (Trusty Tahr)将最低内存需求设为 1GB，Ubuntu 18.04 LTS (Bionic Beaver)提高到 4GB，现在又一次提高了内存需求。相比下，微软将 Windows 11 的最低内存需求设为 4GB，当然这不过是微软的又一个谎言，没人真的会在只有 4GB 内存的计算机运行微软的最新操作系统，Windows 11 至少需要 8GB 内存。Canonical 并没有将 6GB 内存作为硬条件，用户仍然能在不到 6GB 内存的计算机上安装 Ubuntu 26.04。

日本科学家 Yasuto Narukiyo 在 ISSCC 上演示了能承受核反应堆六个月强辐射的 Wi-Fi 接收器。Wi-Fi 接收器能承受 500 kilograys，是太空电子设备通常三年承受 100-300 grays 辐射剂量的千倍以上。Narukiyo 称，2011 年福岛核事故后工程师使用机器人勘察和清理核电站。这些机器人多需要 LAN 线缆，而线缆很容易缠绕在一起。他的团队的目标是开发一种用于在这种恶劣环境下控制机器人的无线系统。即使不是如此极端的环境，核电站寿命到期也需要清理，而很多关闭的核电站至今未完成清理工作，未来二十年还有 200 座反应堆退役。Narukiyo 的团队使用了硅 MOSFET 晶体管，减少晶体管数量同时改变其形状，并加宽了栅极。

由于风能和太阳能发电量创新高，英国在 2026 年 3 月免于进口价值 10 亿英镑的天然气。3 月的风能和太阳能发电总量达到 11TWh，合计增长 28%，英国因此避免了进口 21 TWh 的天然气，按照目前的价格相当于 10 亿英镑。相比 2022 年俄罗斯入侵乌克兰导致 3 月油气价格高涨，中东战事后的天然气价格对英国电价的影响降低了约 25%。

多位流行 NPM 软件包维护者成为 AI 深度伪造攻击目标，他们遭遇了相似的社会工程攻击。axios 维护者 Jason Saayman 称，疑似 APT 组织 UNC1069 的黑客冒充一家公司的创始人联系他，他们不仅克隆了创始人的外表，还克隆了公司本身。他们邀请他加入一个真实的 Slack 工作区（workspace），还创建了频道分享 LinkedIn 帖子，非常逼真。然后黑客邀请参加一个 Microsoft Teams 虚拟会议，会议提示其系统存在问题。他以为这与 Teams 有关，于是安装了缺失组件，结果却植入了远程访问木马。他维护的 axios 周下载量 1 亿，被云服务和编码环境广泛使用，黑客窃取了维护者的凭证释出了 axios 的恶意版本。这不是一起孤立事件，多位周下载量上亿的 NPM 软件包维护者遭遇了类似的 AI 深度伪造攻击。

The Document Foundation（TDF）再次通过官方博客回应了它与主要商业合作伙伴 Collabora 之间的分歧。TDF 称过去几年它犯下了多项违反非营利组织法的错误：仅允许生态系统内的公司免费使用 LibreOffice 品牌；将 LibreOffice 的开发合同——新功能开发、修 bug 等——授予在基金会董事会中拥有代表，积极参与采购的企业。基金会法律顾问指出这些违反法律的错误之后，从中受益的企业试图维持现状而不是解决问题。为避免失去非营利组织地位以及由此带来的不可预见的后果，TDF 基金会采取了措施，取消 Collabora 员工的会员资格、冻结招标以及引入开发采购政策，并制定了规则降低未来再次出现类似问题的风险。

Linux 准备移除对 i486 CPU 的支持，名为“x86/cpu: Remove M486/M486SX/ELAN support”的补丁预计将合并到 Linux 7.1 中。英特尔是在 1989 年 4 月推出了 25MHz i486 CPU，33MHz i486 在 1990 年 5 月发布，50MHz 在 1991 年 6 月推出。英特尔对 i486 的生产一直持续到 2007 年。i486 的继任者是 1993 年推出的奔腾 CPU。i486 也是 AMD 的最后一款英特尔处理器克隆产品。Linux kernel 过去几年已经多次考虑移除对 i486 的支持。Linux 作者 Linus Torvalds 最近再次表示不应该再浪费精力在 i486 的支持上。

《纽约客》发表了一篇长文，从 OpenAI 董事会在 2023 年秋季发起的未遂“政变”谈起，使用了未完整公开的内部备忘录，详述了 OpenAI 受争议 CEO Sam Altman 过去十几年的经历，提出了一个问题：如果今天的大模型能通向 AGI（通用人工智能），那么 OpenAI 及其 CEO 有可能控制人类的未来，但 Sam 能被人信任吗？文章援引了他在 Y Combinator 第一批创业营的同学、2013 年自杀的 Aaron Swartz 的评价，“Sam 永远不可被信任，他是个反社会人格者。什么都做得出来。”微软在 2023 年的未遂“政变”中支持了 Sam，如今双方关系紧张， 微软高管认为他可能会作为一个骗子留在人们的记忆里。