{
  "title": "每日研究简报 2026-04-26",
  "url": "/posts/research-brief-2026-04-26/",
  "permalink": "https://hackcv.com/posts/research-brief-2026-04-26/",
  "date": "2026-04-26",
  "lastmod": "2026-04-26",
  "author": "",
  "description": "AI / 大模型 / Agent / 计算机视觉 / 音视频处理算法 / 工程优化 领域每日研究简报",
  "categories": ["研究简报"],
  "tags": ["AI","大模型","Agent","计算机视觉","音视频处理","工程优化","每日简报"],
  "cover": "https://picsum.photos/seed/%E6%AF%8F%E6%97%A5%E7%A0%94%E7%A9%B6%E7%AE%80%E6%8A%A5-2026-04-26/1200/675",
  "readingTime": 7,
  "wordCount": 2057,
  "content": "\u003cblockquote\u003e\n\u003cp\u003e📅 生成时间：2026-04-26 21:17 (Asia/Shanghai) | 数据来源：arXiv · GitHub · HackerNews · 科技媒体 · 大厂博客\u003c/p\u003e\n\u003c/blockquote\u003e\n\u003chr\u003e\n\u003ch2 id=\"-一arxiv-最新论文\"\u003e📄 一、arXiv 最新论文\u003c/h2\u003e\n\u003ch3 id=\"1-seeing-fast-and-slow-learning-the-flow-of-time-in-videos\"\u003e1. Seeing Fast and Slow: Learning the Flow of Time in Videos\u003c/h3\u003e\n\u003cul\u003e\n\u003cli\u003e\u003cstrong\u003e方向\u003c/strong\u003e：arXiv/CV\u003c/li\u003e\n\u003cli\u003e\u003cstrong\u003e摘要\u003c/strong\u003e：How can we tell whether a video has been sped up or slowed down? How can we generate videos at different speeds? Although videos have been central to modern computer vision research, little attention has been paid to perceiving and controlling the passage of time. In this paper, we study time as a learnable visual concept and develop models for reasoning about and manipulating the flow of time in \u0026hellip;\u003c/li\u003e\n\u003cli\u003e\u003cstrong\u003e推荐原因\u003c/strong\u003e：3D 感知与重建是自动驾驶、AR/VR 的核心技术，工程价值突出。\u003c/li\u003e\n\u003cli\u003e\u003cstrong\u003e链接\u003c/strong\u003e：https://arxiv.org/abs/2604.21931v1\u003c/li\u003e\n\u003c/ul\u003e\n\u003ch3 id=\"2-seeing-without-eyes-4d-human-scene-understanding-from-wearable-imus\"\u003e2. Seeing Without Eyes: 4D Human-Scene Understanding from Wearable IMUs\u003c/h3\u003e\n\u003cul\u003e\n\u003cli\u003e\u003cstrong\u003e方向\u003c/strong\u003e：arXiv/CV\u003c/li\u003e\n\u003cli\u003e\u003cstrong\u003e摘要\u003c/strong\u003e：Understanding human activities and their surrounding environments typically relies on visual perception, yet cameras pose persistent challenges in privacy, safety, energy efficiency, and scalability. We explore an alternative: 4D perception without vision. Its goal is to reconstruct human motion and 3D scene layouts purely from everyday wearable sensors. For this we introduce IMU-to-4D, a framewor\u0026hellip;\u003c/li\u003e\n\u003cli\u003e\u003cstrong\u003e推荐原因\u003c/strong\u003e：模型安全评测是保证大模型可靠部署的基础。\u003c/li\u003e\n\u003cli\u003e\u003cstrong\u003e链接\u003c/strong\u003e：https://arxiv.org/abs/2604.21926v1\u003c/li\u003e\n\u003c/ul\u003e\n\u003ch3 id=\"3-temporal-taskification-in-streaming-continual-learning-a-source-of-evaluation-instability\"\u003e3. Temporal Taskification in Streaming Continual Learning: A Source of Evaluation Instability\u003c/h3\u003e\n\u003cul\u003e\n\u003cli\u003e\u003cstrong\u003e方向\u003c/strong\u003e：arXiv/LG\u003c/li\u003e\n\u003cli\u003e\u003cstrong\u003e摘要\u003c/strong\u003e：Streaming Continual Learning (CL) typically converts a continuous stream into a sequence of discrete tasks through temporal partitioning. We argue that this temporal taskification step is not a neutral preprocessing choice, but a structural component of evaluation: different valid splits of the same stream can induce different CL regimes and therefore different benchmark conclusions. To study this\u0026hellip;\u003c/li\u003e\n\u003cli\u003e\u003cstrong\u003e推荐原因\u003c/strong\u003e：视频 AI 处理（生成、压缩、增强）是下一个 AIGC 增长点。\u003c/li\u003e\n\u003cli\u003e\u003cstrong\u003e链接\u003c/strong\u003e：https://arxiv.org/abs/2604.21930v1\u003c/li\u003e\n\u003c/ul\u003e\n\u003ch3 id=\"4-fine-tuning-regimes-define-distinct-continual-learning-problems\"\u003e4. Fine-Tuning Regimes Define Distinct Continual Learning Problems\u003c/h3\u003e\n\u003cul\u003e\n\u003cli\u003e\u003cstrong\u003e方向\u003c/strong\u003e：arXiv/LG\u003c/li\u003e\n\u003cli\u003e\u003cstrong\u003e摘要\u003c/strong\u003e：Continual learning (CL) studies how models acquire tasks sequentially while retaining previously learned knowledge. Despite substantial progress in benchmarking CL methods, comparative evaluations typically keep the fine-tuning regime fixed. In this paper, we argue that the fine-tuning regime, defined by the trainable parameter subspace, is itself a key evaluation variable. We formalize adaptation\u0026hellip;\u003c/li\u003e\n\u003cli\u003e\u003cstrong\u003e推荐原因\u003c/strong\u003e：模型安全评测是保证大模型可靠部署的基础。\u003c/li\u003e\n\u003cli\u003e\u003cstrong\u003e链接\u003c/strong\u003e：https://arxiv.org/abs/2604.21927v1\u003c/li\u003e\n\u003c/ul\u003e\n\u003ch3 id=\"5-when-prompts-override-vision-prompt-induced-hallucinations-in-lvlms\"\u003e5. When Prompts Override Vision: Prompt-Induced Hallucinations in LVLMs\u003c/h3\u003e\n\u003cul\u003e\n\u003cli\u003e\u003cstrong\u003e方向\u003c/strong\u003e：arXiv/AI\u003c/li\u003e\n\u003cli\u003e\u003cstrong\u003e摘要\u003c/strong\u003e：Despite impressive progress in capabilities of large vision-language models (LVLMs), these systems remain vulnerable to hallucinations, i.e., outputs that are not grounded in the visual input. Prior work has attributed hallucinations in LVLMs to factors such as limitations of the vision backbone or the dominance of the language component, yet the relative importance of these factors remains unclea\u0026hellip;\u003c/li\u003e\n\u003cli\u003e\u003cstrong\u003e推荐原因\u003c/strong\u003e：内容偏向工程实践，对实际项目有一定帮助。\u003c/li\u003e\n\u003cli\u003e\u003cstrong\u003e链接\u003c/strong\u003e：https://arxiv.org/abs/2604.21911v1\u003c/li\u003e\n\u003c/ul\u003e\n\u003ch3 id=\"6-from-research-question-to-scientific-workflow-leveraging-agentic-ai-for-science-automation\"\u003e6. From Research Question to Scientific Workflow: Leveraging Agentic AI for Science Automation\u003c/h3\u003e\n\u003cul\u003e\n\u003cli\u003e\u003cstrong\u003e方向\u003c/strong\u003e：arXiv/AI\u003c/li\u003e\n\u003cli\u003e\u003cstrong\u003e摘要\u003c/strong\u003e：Scientific workflow systems automate execution \u0026ndash; scheduling, fault tolerance, resource management \u0026ndash; but not the semantic translation that precedes it. Scientists still manually convert research questions into workflow specifications, a task requiring both domain knowledge and infrastructure expertise. We propose an agentic architecture that closes this gap through three layers: an LLM interprets\u0026hellip;\u003c/li\u003e\n\u003cli\u003e\u003cstrong\u003e推荐原因\u003c/strong\u003e：多 Agent 协作是今年最活跃的研究方向之一，展示了 AI 系统自动化的新可能。\u003c/li\u003e\n\u003cli\u003e\u003cstrong\u003e链接\u003c/strong\u003e：https://arxiv.org/abs/2604.21910v1\u003c/li\u003e\n\u003c/ul\u003e\n\u003ch3 id=\"7-evaluation-of-automatic-speech-recognition-using-generative-large-language-models\"\u003e7. Evaluation of Automatic Speech Recognition Using Generative Large Language Models\u003c/h3\u003e\n\u003cul\u003e\n\u003cli\u003e\u003cstrong\u003e方向\u003c/strong\u003e：arXiv/CL\u003c/li\u003e\n\u003cli\u003e\u003cstrong\u003e摘要\u003c/strong\u003e：Automatic Speech Recognition (ASR) is traditionally evaluated using Word Error Rate (WER), a metric that is insensitive to meaning. Embedding-based semantic metrics are better correlated with human perception, but decoder-based Large Language Models (LLMs) remain underexplored for this task. This paper evaluates their relevance through three approaches: (1) selecting the best hypothesis between tw\u0026hellip;\u003c/li\u003e\n\u003cli\u003e\u003cstrong\u003e推荐原因\u003c/strong\u003e：LLM 的工程优化和推理加速是产业落地的关键瓶颈，持续有新方案涌现。\u003c/li\u003e\n\u003cli\u003e\u003cstrong\u003e链接\u003c/strong\u003e：https://arxiv.org/abs/2604.21928v1\u003c/li\u003e\n\u003c/ul\u003e\n\u003ch3 id=\"8-mathduels-evaluating-llms-as-problem-posers-and-solvers\"\u003e8. MathDuels: Evaluating LLMs as Problem Posers and Solvers\u003c/h3\u003e\n\u003cul\u003e\n\u003cli\u003e\u003cstrong\u003e方向\u003c/strong\u003e：arXiv/CL\u003c/li\u003e\n\u003cli\u003e\u003cstrong\u003e摘要\u003c/strong\u003e：As frontier language models attain near-ceiling performance on static mathematical benchmarks, existing evaluations are increasingly unable to differentiate model capabilities, largely because they cast models solely as solvers of fixed problem sets. We introduce MathDuels, a self-play benchmark in which models occupy dual roles: each authors math problems under adversarial prompting and solves pr\u0026hellip;\u003c/li\u003e\n\u003cli\u003e\u003cstrong\u003e推荐原因\u003c/strong\u003e：AI 安全和对齐问题日益突出，评估体系和防护手段是重要研究方向。\u003c/li\u003e\n\u003cli\u003e\u003cstrong\u003e链接\u003c/strong\u003e：https://arxiv.org/abs/2604.21916v1\u003c/li\u003e\n\u003c/ul\u003e\n\u003ch2 id=\"-二github-热门项目\"\u003e🌟 二、GitHub 热门项目\u003c/h2\u003e\n\u003ch3 id=\"1-alchaincyfhuashu-design\"\u003e1. alchaincyf/huashu-design\u003c/h3\u003e\n\u003cul\u003e\n\u003cli\u003e\u003cstrong\u003eStars\u003c/strong\u003e：⭐ 6,796 · HTML\u003c/li\u003e\n\u003cli\u003e\u003cstrong\u003e简介\u003c/strong\u003e：Huashu Design · HTML-native design skill for Claude Code · Claude Code 里 HTML 原生的设计 skill · 高保真原型 / 幻灯片 / 动画 + 20 设计哲学 + 5 维评审 + MP4 导出 · Agent-agnostic\u003c/li\u003e\n\u003cli\u003e\u003cstrong\u003e推荐原因\u003c/strong\u003e：Web/代码 Agent 落地加速，工程实践价值显著。\u003c/li\u003e\n\u003cli\u003e\u003cstrong\u003e链接\u003c/strong\u003e：https://github.com/alchaincyf/huashu-design\u003c/li\u003e\n\u003c/ul\u003e\n\u003ch3 id=\"2-conardligarden-skills\"\u003e2. ConardLi/garden-skills\u003c/h3\u003e\n\u003cul\u003e\n\u003cli\u003e\u003cstrong\u003eStars\u003c/strong\u003e：⭐ 1,324 · JavaScript\u003c/li\u003e\n\u003cli\u003e\u003cstrong\u003e简介\u003c/strong\u003e：ConardLi\u0026rsquo;s open-source Skills collection, featuring web design, knowledge retrieval, image generation, and more.\u003c/li\u003e\n\u003cli\u003e\u003cstrong\u003e推荐原因\u003c/strong\u003e：AI 安全和对齐问题日益突出，评估体系和防护手段是重要研究方向。\u003c/li\u003e\n\u003cli\u003e\u003cstrong\u003e链接\u003c/strong\u003e：https://github.com/ConardLi/garden-skills\u003c/li\u003e\n\u003c/ul\u003e\n\u003ch3 id=\"3-cosmicstack-labsmercury-agent\"\u003e3. cosmicstack-labs/mercury-agent\u003c/h3\u003e\n\u003cul\u003e\n\u003cli\u003e\u003cstrong\u003eStars\u003c/strong\u003e：⭐ 1,297 · TypeScript\u003c/li\u003e\n\u003cli\u003e\u003cstrong\u003e简介\u003c/strong\u003e：Soul-driven AI agent with permission-hardened tools, token budgets, and multi-channel access. Runs 24/7 from CLI or Telegram.\u003c/li\u003e\n\u003cli\u003e\u003cstrong\u003e推荐原因\u003c/strong\u003e：多 Agent 协作是今年最活跃的研究方向之一，展示了 AI 系统自动化的新可能。\u003c/li\u003e\n\u003cli\u003e\u003cstrong\u003e链接\u003c/strong\u003e：https://github.com/cosmicstack-labs/mercury-agent\u003c/li\u003e\n\u003c/ul\u003e\n\u003ch3 id=\"4-leigest519opengame\"\u003e4. leigest519/OpenGame\u003c/h3\u003e\n\u003cul\u003e\n\u003cli\u003e\u003cstrong\u003eStars\u003c/strong\u003e：⭐ 1,187 · TypeScript\u003c/li\u003e\n\u003cli\u003e\u003cstrong\u003e简介\u003c/strong\u003e：OpenGame: Open Agentic Coding for Games\u003c/li\u003e\n\u003cli\u003e\u003cstrong\u003e推荐原因\u003c/strong\u003e：多 Agent 协作是今年最活跃的研究方向之一，展示了 AI 系统自动化的新可能。\u003c/li\u003e\n\u003cli\u003e\u003cstrong\u003e链接\u003c/strong\u003e：https://github.com/leigest519/OpenGame\u003c/li\u003e\n\u003c/ul\u003e\n\u003ch3 id=\"5-gammalabtechnologiesharmonist\"\u003e5. GammaLabTechnologies/harmonist\u003c/h3\u003e\n\u003cul\u003e\n\u003cli\u003e\u003cstrong\u003eStars\u003c/strong\u003e：⭐ 675 · Python\u003c/li\u003e\n\u003cli\u003e\u003cstrong\u003e简介\u003c/strong\u003e：Portable AI agent orchestration with mechanical protocol enforcement. 186 agents, zero runtime dependencies.\u003c/li\u003e\n\u003cli\u003e\u003cstrong\u003e推荐原因\u003c/strong\u003e：Agent 是当前 AI 工程化的核心方向，代表了大模型从「对话」到「执行」的关键跃迁。\u003c/li\u003e\n\u003cli\u003e\u003cstrong\u003e链接\u003c/strong\u003e：https://github.com/GammaLabTechnologies/harmonist\u003c/li\u003e\n\u003c/ul\u003e\n\u003ch3 id=\"6-zeroz-labcc-design\"\u003e6. ZeroZ-lab/cc-design\u003c/h3\u003e\n\u003cul\u003e\n\u003cli\u003e\u003cstrong\u003eStars\u003c/strong\u003e：⭐ 634 · JavaScript\u003c/li\u003e\n\u003cli\u003e\u003cstrong\u003e简介\u003c/strong\u003e：High-fidelity HTML design and prototype guidance skill for AI agents\u003c/li\u003e\n\u003cli\u003e\u003cstrong\u003e推荐原因\u003c/strong\u003e：引发了不少讨论，值得了解业界观点。\u003c/li\u003e\n\u003cli\u003e\u003cstrong\u003e链接\u003c/strong\u003e：https://github.com/ZeroZ-lab/cc-design\u003c/li\u003e\n\u003c/ul\u003e\n\u003ch3 id=\"7-earthtojaketext-to-cad\"\u003e7. earthtojake/text-to-cad\u003c/h3\u003e\n\u003cul\u003e\n\u003cli\u003e\u003cstrong\u003eStars\u003c/strong\u003e：⭐ 547 · JavaScript\u003c/li\u003e\n\u003cli\u003e\u003cstrong\u003e简介\u003c/strong\u003e：An open source harness for generating CAD models\u003c/li\u003e\n\u003cli\u003e\u003cstrong\u003e推荐原因\u003c/strong\u003e：从社区反馈来看有一定价值，可以快速浏览。\u003c/li\u003e\n\u003cli\u003e\u003cstrong\u003e链接\u003c/strong\u003e：https://github.com/earthtojake/text-to-cad\u003c/li\u003e\n\u003c/ul\u003e\n\u003ch3 id=\"8-wuyoscargpt_image_2_skill\"\u003e8. wuyoscar/gpt_image_2_skill\u003c/h3\u003e\n\u003cul\u003e\n\u003cli\u003e\u003cstrong\u003eStars\u003c/strong\u003e：⭐ 527 · Python\u003c/li\u003e\n\u003cli\u003e\u003cstrong\u003e简介\u003c/strong\u003e：GPT Image 2 prompt gallery, image prompt library, agentic skill, and CLI for OpenAI image generation/editing\u003c/li\u003e\n\u003cli\u003e\u003cstrong\u003e推荐原因\u003c/strong\u003e：Agent 是当前 AI 工程化的核心方向，代表了大模型从「对话」到「执行」的关键跃迁。\u003c/li\u003e\n\u003cli\u003e\u003cstrong\u003e链接\u003c/strong\u003e：https://github.com/wuyoscar/gpt_image_2_skill\u003c/li\u003e\n\u003c/ul\u003e\n\u003ch2 id=\"-三ai-科技媒体--大厂博客\"\u003e📰 三、AI 科技媒体 \u0026amp; 大厂博客\u003c/h2\u003e\n\u003ch3 id=\"1-anthropic-created-a-test-marketplace-for-agent-on-agent-commerce\"\u003e1. Anthropic created a test marketplace for agent-on-agent commerce\u003c/h3\u003e\n\u003cul\u003e\n\u003cli\u003e\u003cstrong\u003e来源\u003c/strong\u003e：TechCrunch AI · AI 媒体\u003c/li\u003e\n\u003cli\u003e\u003cstrong\u003e摘要\u003c/strong\u003e：In a recent experiment, Anthropic created a classified marketplace where AI agents represented both buyers and sellers, striking real deals for real goods and real money.\u003c/li\u003e\n\u003cli\u003e\u003cstrong\u003e推荐原因\u003c/strong\u003e：多 Agent 协作是今年最活跃的研究方向之一，展示了 AI 系统自动化的新可能。\u003c/li\u003e\n\u003cli\u003e\u003cstrong\u003e链接\u003c/strong\u003e：https://techcrunch.com/2026/04/25/anthropic-created-a-test-marketplace-for-agent-on-agent-commerce/\u003c/li\u003e\n\u003c/ul\u003e\n\u003ch3 id=\"2-maines-governor-vetoes-data-center-moratorium\"\u003e2. Maine’s governor vetoes data center moratorium\u003c/h3\u003e\n\u003cul\u003e\n\u003cli\u003e\u003cstrong\u003e来源\u003c/strong\u003e：TechCrunch AI · AI 媒体\u003c/li\u003e\n\u003cli\u003e\u003cstrong\u003e摘要\u003c/strong\u003e：L.D. 307 would have imposed the country’s first statewide moratorium on new data centers — lasting, in this case, until November 1, 2027.\u003c/li\u003e\n\u003cli\u003e\u003cstrong\u003e推荐原因\u003c/strong\u003e：引发了不少讨论，值得了解业界观点。\u003c/li\u003e\n\u003cli\u003e\u003cstrong\u003e链接\u003c/strong\u003e：https://techcrunch.com/2026/04/25/maines-governor-vetoes-data-center-moratorium/\u003c/li\u003e\n\u003c/ul\u003e\n\u003ch3 id=\"3-openai-ceo-apologizes-to-tumbler-ridge-community\"\u003e3. OpenAI CEO apologizes to Tumbler Ridge community\u003c/h3\u003e\n\u003cul\u003e\n\u003cli\u003e\u003cstrong\u003e来源\u003c/strong\u003e：TechCrunch AI · AI 媒体\u003c/li\u003e\n\u003cli\u003e\u003cstrong\u003e摘要\u003c/strong\u003e：In a letter to the residents of Tumbler Ridge, Canada, OpenAI CEO Sam Altman said he is “deeply sorry” that his company failed to alert law enforcement about the suspect in a recent mass shooting.\u003c/li\u003e\n\u003cli\u003e\u003cstrong\u003e推荐原因\u003c/strong\u003e：提供了一些新的思路和视角，可以扩展知识面。\u003c/li\u003e\n\u003cli\u003e\u003cstrong\u003e链接\u003c/strong\u003e：https://techcrunch.com/2026/04/25/openai-ceo-apologizes-to-tumbler-ridge-community/\u003c/li\u003e\n\u003c/ul\u003e\n\u003ch3 id=\"4-discord-sleuths-gained-unauthorized-access-to-anthropics-mythos\"\u003e4. Discord Sleuths Gained Unauthorized Access to Anthropic’s Mythos\u003c/h3\u003e\n\u003cul\u003e\n\u003cli\u003e\u003cstrong\u003e来源\u003c/strong\u003e：Wired AI · AI 媒体\u003c/li\u003e\n\u003cli\u003e\u003cstrong\u003e摘要\u003c/strong\u003e：Plus: Spy firms tap into a global telecom weakness to track targets, 500,000 UK health records go up for sale on Alibaba, Apple patches a revealing notification bug, and more.\u003c/li\u003e\n\u003cli\u003e\u003cstrong\u003e推荐原因\u003c/strong\u003e：内容偏向工程实践，对实际项目有一定帮助。\u003c/li\u003e\n\u003cli\u003e\u003cstrong\u003e链接\u003c/strong\u003e：https://www.wired.com/story/security-news-this-week-discord-sleuths-gained-unauthorized-access-to-anthropics-mythos/\u003c/li\u003e\n\u003c/ul\u003e\n\u003ch3 id=\"5-ace-the-ping-pong-robot-can-whup-your-ass\"\u003e5. Ace the Ping-Pong Robot Can Whup Your Ass\u003c/h3\u003e\n\u003cul\u003e\n\u003cli\u003e\u003cstrong\u003e来源\u003c/strong\u003e：Wired AI · AI 媒体\u003c/li\u003e\n\u003cli\u003e\u003cstrong\u003e摘要\u003c/strong\u003e：Ace can read the trajectory of a ball, adjust the racket angle, and respond with strokes that keep the exchange alive with real players.\u003c/li\u003e\n\u003cli\u003e\u003cstrong\u003e推荐原因\u003c/strong\u003e：HN 社区讨论热度高，反映了开发者社区的真实关注点。\u003c/li\u003e\n\u003cli\u003e\u003cstrong\u003e链接\u003c/strong\u003e：https://www.wired.com/story/ace-the-robot-wants-to-become-the-world-table-tennis-champion/\u003c/li\u003e\n\u003c/ul\u003e\n\u003ch3 id=\"6-ai-designed-drugs-by-a-deepmind-spinoff-are-headed-to-human-trials\"\u003e6. AI-Designed Drugs by a DeepMind Spinoff Are Headed to Human Trials\u003c/h3\u003e\n\u003cul\u003e\n\u003cli\u003e\u003cstrong\u003e来源\u003c/strong\u003e：Wired AI · AI 媒体\u003c/li\u003e\n\u003cli\u003e\u003cstrong\u003e摘要\u003c/strong\u003e：Isomorphic Labs president Max Jaderberg said at WIRED Health in London that the startup has built a “broad and exciting pipeline of new medicines.”\u003c/li\u003e\n\u003cli\u003e\u003cstrong\u003e推荐原因\u003c/strong\u003e：内容偏向工程实践，对实际项目有一定帮助。\u003c/li\u003e\n\u003cli\u003e\u003cstrong\u003e链接\u003c/strong\u003e：https://www.wired.com/story/wired-health-2026-how-ai-is-powering-drug-discovery-max-jaderberg/\u003c/li\u003e\n\u003c/ul\u003e\n\u003ch3 id=\"7-why-are-top-university-websites-serving-porn-it-comes-down-to-shoddy-housekeeping\"\u003e7. Why are top university websites serving porn? It comes down to shoddy housekeeping.\u003c/h3\u003e\n\u003cul\u003e\n\u003cli\u003e\u003cstrong\u003e来源\u003c/strong\u003e：Ars Technica · AI 媒体\u003c/li\u003e\n\u003cli\u003e\u003cstrong\u003e摘要\u003c/strong\u003e：Hundreds of subdomains from dozens of universities have been hijacked by scammers.\u003c/li\u003e\n\u003cli\u003e\u003cstrong\u003e推荐原因\u003c/strong\u003e：引发了不少讨论，值得了解业界观点。\u003c/li\u003e\n\u003cli\u003e\u003cstrong\u003e链接\u003c/strong\u003e：https://arstechnica.com/security/2026/04/why-are-top-university-websites-serving-porn-it-comes-down-to-shoddy-housekeeping/\u003c/li\u003e\n\u003c/ul\u003e\n\u003ch3 id=\"8-in-a-first-a-ransomware-family-is-confirmed-to-be-quantum-safe\"\u003e8. In a first, a ransomware family is confirmed to be quantum-safe\u003c/h3\u003e\n\u003cul\u003e\n\u003cli\u003e\u003cstrong\u003e来源\u003c/strong\u003e：Ars Technica · AI 媒体\u003c/li\u003e\n\u003cli\u003e\u003cstrong\u003e摘要\u003c/strong\u003e：Technically speaking, there\u0026rsquo;s no practical benefit to use PQC. So why is it being used?\u003c/li\u003e\n\u003cli\u003e\u003cstrong\u003e推荐原因\u003c/strong\u003e：AI 安全和对齐问题日益突出，评估体系和防护手段是重要研究方向。\u003c/li\u003e\n\u003cli\u003e\u003cstrong\u003e链接\u003c/strong\u003e：https://arstechnica.com/security/2026/04/now-even-ransomware-is-using-post-quantum-cryptography/\u003c/li\u003e\n\u003c/ul\u003e\n\u003ch3 id=\"9-microsoft-issues-emergency-update-for-macos-and-linux-aspnet-threat\"\u003e9. Microsoft issues emergency update for macOS and Linux ASP.NET threat\u003c/h3\u003e\n\u003cul\u003e\n\u003cli\u003e\u003cstrong\u003e来源\u003c/strong\u003e：Ars Technica · AI 媒体\u003c/li\u003e\n\u003cli\u003e\u003cstrong\u003e摘要\u003c/strong\u003e：When authentication fails, things can go very, very wrong.\u003c/li\u003e\n\u003cli\u003e\u003cstrong\u003e推荐原因\u003c/strong\u003e：是近期技术趋势的典型案例，有一定代表性。\u003c/li\u003e\n\u003cli\u003e\u003cstrong\u003e链接\u003c/strong\u003e：https://arstechnica.com/security/2026/04/microsoft-issues-emergency-update-for-macos-and-linux-asp-net-threat/\u003c/li\u003e\n\u003c/ul\u003e\n\u003ch3 id=\"10-three-reasons-why-deepseeks-new-model-matters\"\u003e10. Three reasons why DeepSeek’s new model matters\u003c/h3\u003e\n\u003cul\u003e\n\u003cli\u003e\u003cstrong\u003e来源\u003c/strong\u003e：MIT Tech Review · AI 媒体\u003c/li\u003e\n\u003cli\u003e\u003cstrong\u003e摘要\u003c/strong\u003e：On Friday, Chinese AI firm DeepSeek released a preview of V4, its long-awaited new flagship model. Notably, the model can process much longer prompts than its last generation, thanks to a new design t\u003c/li\u003e\n\u003cli\u003e\u003cstrong\u003e推荐原因\u003c/strong\u003e：是近期技术趋势的典型案例，有一定代表性。\u003c/li\u003e\n\u003cli\u003e\u003cstrong\u003e链接\u003c/strong\u003e：https://www.technologyreview.com/2026/04/24/1136422/why-deepseeks-v4-matters/\u003c/li\u003e\n\u003c/ul\u003e\n\u003ch3 id=\"11-the-download-supercharged-scams-and-studying-ai-healthcare\"\u003e11. The Download: supercharged scams and studying AI healthcare\u003c/h3\u003e\n\u003cul\u003e\n\u003cli\u003e\u003cstrong\u003e来源\u003c/strong\u003e：MIT Tech Review · AI 媒体\u003c/li\u003e\n\u003cli\u003e\u003cstrong\u003e摘要\u003c/strong\u003e：This is today’s edition of The Download, our weekday newsletter that provides a daily dose of what’s going on in the world of technology. We’re in a new era of AI-driven scams When ChatGPT was release\u003c/li\u003e\n\u003cli\u003e\u003cstrong\u003e推荐原因\u003c/strong\u003e：从社区反馈来看有一定价值，可以快速浏览。\u003c/li\u003e\n\u003cli\u003e\u003cstrong\u003e链接\u003c/strong\u003e：https://www.technologyreview.com/2026/04/24/1136400/the-download-supercharged-scams-questionable-ai-healthcare/\u003c/li\u003e\n\u003c/ul\u003e\n\u003ch3 id=\"12-health-care-ai-is-here-we-dont-know-if-it-actually-helps-patients\"\u003e12. Health-care AI is here. We don’t know if it actually helps patients.\u003c/h3\u003e\n\u003cul\u003e\n\u003cli\u003e\u003cstrong\u003e来源\u003c/strong\u003e：MIT Tech Review · AI 媒体\u003c/li\u003e\n\u003cli\u003e\u003cstrong\u003e摘要\u003c/strong\u003e：I don’t need to tell you that AI is everywhere. Or that it is being used, increasingly, in hospitals. Doctors are using AI to help them with notetaking. AI-based tools are trawling through patient rec\u003c/li\u003e\n\u003cli\u003e\u003cstrong\u003e推荐原因\u003c/strong\u003e：技术实现有一定参考价值，可借鉴到类似项目中。\u003c/li\u003e\n\u003cli\u003e\u003cstrong\u003e链接\u003c/strong\u003e：https://www.technologyreview.com/2026/04/24/1136352/health-care-ai-dont-know-actually-helps-patients/\u003c/li\u003e\n\u003c/ul\u003e\n\u003ch2 id=\"-四hackernews-近-48h-热门\"\u003e🔥 四、HackerNews 近 48h 热门\u003c/h2\u003e\n\u003ch3 id=\"1-i-cancelled-claude-token-issues-declining-quality-and-poor-support\"\u003e1. I cancelled Claude: Token issues, declining quality, and poor support\u003c/h3\u003e\n\u003cul\u003e\n\u003cli\u003e\u003cstrong\u003e热度\u003c/strong\u003e：950 points · 💬 567 comments\u003c/li\u003e\n\u003cli\u003e\u003cstrong\u003e推荐原因\u003c/strong\u003e：HN 热门文章，热度很高，强烈推荐。\u003c/li\u003e\n\u003cli\u003e\u003cstrong\u003e链接\u003c/strong\u003e：https://nickyreinert.de/en/2026/2026-04-24-claude-critics/\u003c/li\u003e\n\u003cli\u003e\u003cstrong\u003eHN 讨论\u003c/strong\u003e：https://news.ycombinator.com/item?id=47892019\u003c/li\u003e\n\u003cli\u003e\u003cstrong\u003e高赞评论（原文+中文）\u003c/strong\u003e：\n· \u003cstrong\u003ewg0\u003c/strong\u003e ：I write detailed specs. Multifile with example code. In markdown. Then hand over to Claude Sonnet. With hard requirements listed, I found out that the generated code missed requirements, had duplicate code or even unnecessary code wrangling data (mapping objects into new objects of narrower types…\n→ 我写详细的规格。包含示例代码的多文件。扣分后，然后交给克劳德·索奈特。列出硬要求后，我发现生成的代码错过了要求，有重复的代码，甚至有不必要的代码争吵数据（将对象映射到狭窄类型的新对象……\n· \u003cstrong\u003erectang\u003c/strong\u003e ：I feel like I\u0026rsquo;m using Claude Opus pretty effectively and I\u0026rsquo;m honestly not running up against limits in my mid-tier subscriptions. My workflow is more \u0026ldquo;copilot\u0026rdquo; than \u0026ldquo;autopilot\u0026rdquo;, in that I craft prompts for contained tasks and review nearly everything, so it\u0026rsquo;s pretty light compared to people doing…\n→ 我觉得我非常有效地使用Claude Opus ，老实说，我的中端订阅没有遇到限制。我的工作流程更像是“副驾驶” ，而不是“自动驾驶” ，因为我为包含的任务创建提示并查看几乎所有内容，因此与执行以下操作的人相比，它非常轻松……\n· \u003cstrong\u003ejanwillemb\u003c/strong\u003e ：This is what worries me. People become dependent on these GenAI products that are proprietary, not transparant, and need a subscription. People build on it like it is a solid foundation. But all of a sudden the owner just pulls the foundation from under your building.\n→ 这正是我所担心的。人们变得依赖这些专有而非透明的GenAI产品，需要订阅。人们以此为基础，仿佛它是一个坚实的基础。但突然之间，房东就把地基从您的大楼下拉了出来。\u003c/li\u003e\n\u003c/ul\u003e\n\u003ch3 id=\"2-google-plans-to-invest-up-to-40b-in-anthropic\"\u003e2. Google plans to invest up to $40B in Anthropic\u003c/h3\u003e\n\u003cul\u003e\n\u003cli\u003e\u003cstrong\u003e热度\u003c/strong\u003e：805 points · 💬 811 comments\u003c/li\u003e\n\u003cli\u003e\u003cstrong\u003e推荐原因\u003c/strong\u003e：HN 热门文章，热度很高，强烈推荐。\u003c/li\u003e\n\u003cli\u003e\u003cstrong\u003e链接\u003c/strong\u003e：https://www.bloomberg.com/news/articles/2026-04-24/google-plans-to-invest-up-to-40-billion-in-anthropic\u003c/li\u003e\n\u003cli\u003e\u003cstrong\u003eHN 讨论\u003c/strong\u003e：https://news.ycombinator.com/item?id=47892074\u003c/li\u003e\n\u003cli\u003e\u003cstrong\u003e高赞评论（原文+中文）\u003c/strong\u003e：\n· \u003cstrong\u003eelffjs\u003c/strong\u003e ：https://archive.ph/u274V\n· \u003cstrong\u003eskybrian\u003c/strong\u003e ：Context: a few weeks ago, Anthropic signed a deal to buy \u0026ldquo;multiple gigawatts of next-generation TPU capacity\u0026rdquo; from Google and Broadcom [1]. There have been several previous deals, too. Some people call this sort of thing a \u0026ldquo;circular deal\u0026rdquo;, but perhaps a better way to think of it is as a very…\n→ 背景：几周前， Anthropic签署了一项协议，从Google和Broadcom购买“数千兆瓦的下一代TPU容量” [1]。之前也有几笔交易。有些人称这种事情为“循环交易” ，但也许更好的方式是将其视为非常……\n· \u003cstrong\u003e33MHz-i486\u003c/strong\u003e ：I think the subtext of the last few weeks is the Anthropic was becoming severely capacity constrained (or approaching that). They seem to have had to sign two somewhat adverse contracts with Amazon and Google in short succession. suddenly model quality is back up again.\n→ 我认为过去几周的潜台词是人类正在变得严重的能力限制（或接近）。他们似乎不得不在短时间内与亚马逊和谷歌签订了两份有些不利的合同。突然，车型质量又恢复了。\u003c/li\u003e\n\u003c/ul\u003e\n\u003ch3 id=\"3-the-west-forgot-how-to-make-things-now-its-forgetting-how-to-code\"\u003e3. The West forgot how to make things, now it’s forgetting how to code\u003c/h3\u003e\n\u003cul\u003e\n\u003cli\u003e\u003cstrong\u003e热度\u003c/strong\u003e：598 points · 💬 359 comments\u003c/li\u003e\n\u003cli\u003e\u003cstrong\u003e推荐原因\u003c/strong\u003e：HN 热门文章，热度很高，强烈推荐。\u003c/li\u003e\n\u003cli\u003e\u003cstrong\u003e链接\u003c/strong\u003e：https://techtrenches.dev/p/the-west-forgot-how-to-make-things\u003c/li\u003e\n\u003cli\u003e\u003cstrong\u003eHN 讨论\u003c/strong\u003e：https://news.ycombinator.com/item?id=47907879\u003c/li\u003e\n\u003cli\u003e\u003cstrong\u003e高赞评论（原文+中文）\u003c/strong\u003e：\n· \u003cstrong\u003ejdw64\u003c/strong\u003e ：The real issue, in my view, is not AI itself. The problem is a management pattern: removing people and organizational slack because they don’t generate immediate profit, and then expecting the knowledge to still be there when it’s needed. Short-term cost cutting leads to less junior hiring, and…\n→ 在我看来，真正的问题不是人工智能本身。问题在于管理模式：消除人员和组织松懈，因为他们不会立即产生利润，然后期望知识在需要时仍然存在。短期成本削减导致初级招聘人数减少，而且……\n· \u003cstrong\u003eliendolucas\u003c/strong\u003e ：I still code daily without any coding assistance mostly because I believe this is the way to not forget how things are done, even trivial things. My main point against using AI is that I do not want to depend basically on anything when I\u0026rsquo;m in front of the screen (obviously not including,…\n→ 我仍然每天在没有任何编码帮助的情况下进行编码，主要是因为我相信这是一种不会忘记事情是如何完成的，即使是微不足道的事情。我反对使用人工智能的主要观点是，当我在屏幕前时，我不想依赖任何东西（显然不包括， ……\n· \u003cstrong\u003eTonyAlicea10\u003c/strong\u003e ：“Money was never the constraint. Knowledge was.” The irony is how difficult it is to read this obviously AI-generated article due to its unnatural prose and choppy flow full of LLM-isms. The ability to write is also a skill that atrophies. Even when AI is understandably used due to language…\n→ “金钱从来不是制约因素。知识是。”具有讽刺意味的是，由于其不自然的散文和充满LLM主义的波动流动，阅读这篇明显由人工智能生成的文章是多么困难。写作能力也是一种萎缩的技能。即使人工智能因语言而被使用，也是可以理解的……\u003c/li\u003e\n\u003c/ul\u003e\n\u003ch3 id=\"4-new-10-gbe-usb-adapters-are-cooler-smaller-cheaper\"\u003e4. New 10 GbE USB adapters are cooler, smaller, cheaper\u003c/h3\u003e\n\u003cul\u003e\n\u003cli\u003e\u003cstrong\u003e热度\u003c/strong\u003e：585 points · 💬 348 comments\u003c/li\u003e\n\u003cli\u003e\u003cstrong\u003e推荐原因\u003c/strong\u003e：HN 热门文章，热度很高，强烈推荐。\u003c/li\u003e\n\u003cli\u003e\u003cstrong\u003e链接\u003c/strong\u003e：https://www.jeffgeerling.com/blog/2026/new-10-gbe-usb-adapters-cooler-smaller-cheaper/\u003c/li\u003e\n\u003cli\u003e\u003cstrong\u003eHN 讨论\u003c/strong\u003e：https://news.ycombinator.com/item?id=47899053\u003c/li\u003e\n\u003c/ul\u003e\n\u003ch3 id=\"5-sabotaging-projects-by-overthinking-scope-creep-and-structural-diffing\"\u003e5. Sabotaging projects by overthinking, scope creep, and structural diffing\u003c/h3\u003e\n\u003cul\u003e\n\u003cli\u003e\u003cstrong\u003e热度\u003c/strong\u003e：521 points · 💬 135 comments\u003c/li\u003e\n\u003cli\u003e\u003cstrong\u003e推荐原因\u003c/strong\u003e：HN 热门文章，热度很高，强烈推荐。\u003c/li\u003e\n\u003cli\u003e\u003cstrong\u003e链接\u003c/strong\u003e：https://kevinlynagh.com/newsletter/2026_04_overthinking/\u003c/li\u003e\n\u003cli\u003e\u003cstrong\u003eHN 讨论\u003c/strong\u003e：https://news.ycombinator.com/item?id=47890799\u003c/li\u003e\n\u003c/ul\u003e\n\u003ch3 id=\"6-amateur-armed-with-chatgpt-solves-an-erdős-problem\"\u003e6. Amateur armed with ChatGPT solves an Erdős problem\u003c/h3\u003e\n\u003cul\u003e\n\u003cli\u003e\u003cstrong\u003e热度\u003c/strong\u003e：478 points · 💬 309 comments\u003c/li\u003e\n\u003cli\u003e\u003cstrong\u003e推荐原因\u003c/strong\u003e：HN 讨论热烈（309 条评论），社区关注度高。\u003c/li\u003e\n\u003cli\u003e\u003cstrong\u003e链接\u003c/strong\u003e：https://www.scientificamerican.com/article/amateur-armed-with-chatgpt-vibe-maths-a-60-year-old-problem/\u003c/li\u003e\n\u003cli\u003e\u003cstrong\u003eHN 讨论\u003c/strong\u003e：https://news.ycombinator.com/item?id=47903126\u003c/li\u003e\n\u003c/ul\u003e\n\u003ch3 id=\"7-firefox-has-integrated-braves-adblock-engine\"\u003e7. Firefox Has Integrated Brave\u0026rsquo;s Adblock Engine\u003c/h3\u003e\n\u003cul\u003e\n\u003cli\u003e\u003cstrong\u003e热度\u003c/strong\u003e：394 points · 💬 231 comments\u003c/li\u003e\n\u003cli\u003e\u003cstrong\u003e推荐原因\u003c/strong\u003e：HN 讨论热烈（231 条评论），社区关注度高。\u003c/li\u003e\n\u003cli\u003e\u003cstrong\u003e链接\u003c/strong\u003e：https://itsfoss.com/news/firefox-ships-brave-adblock-engine/\u003c/li\u003e\n\u003cli\u003e\u003cstrong\u003eHN 讨论\u003c/strong\u003e：https://news.ycombinator.com/item?id=47897891\u003c/li\u003e\n\u003c/ul\u003e\n\u003ch3 id=\"8-usb-cheat-sheet-2022\"\u003e8. USB Cheat Sheet (2022)\u003c/h3\u003e\n\u003cul\u003e\n\u003cli\u003e\u003cstrong\u003e热度\u003c/strong\u003e：377 points · 💬 66 comments\u003c/li\u003e\n\u003cli\u003e\u003cstrong\u003e推荐原因\u003c/strong\u003e：HN 获得较多关注，质量和讨论度不错。\u003c/li\u003e\n\u003cli\u003e\u003cstrong\u003e链接\u003c/strong\u003e：https://fabiensanglard.net/usbcheat/index.html\u003c/li\u003e\n\u003cli\u003e\u003cstrong\u003eHN 讨论\u003c/strong\u003e：https://news.ycombinator.com/item?id=47904876\u003c/li\u003e\n\u003c/ul\u003e\n\u003ch2 id=\"-深读推荐\"\u003e📚 深读推荐\u003c/h2\u003e\n\u003ctable\u003e\n\u003cthead\u003e\n\u003ctr\u003e\n\u003cth\u003e类型\u003c/th\u003e\n\u003cth\u003e标题\u003c/th\u003e\n\u003cth\u003e方向\u003c/th\u003e\n\u003cth\u003e备注\u003c/th\u003e\n\u003cth\u003e链接\u003c/th\u003e\n\u003c/tr\u003e\n\u003c/thead\u003e\n\u003ctbody\u003e\n\u003ctr\u003e\n\u003ctd\u003e📄 论文\u003c/td\u003e\n\u003ctd\u003eSeeing Fast and Slow: Learning the Flo…\u003c/td\u003e\n\u003ctd\u003eCV\u003c/td\u003e\n\u003ctd\u003e\u003c/td\u003e\n\u003ctd\u003e\u003ca href=\"https://arxiv.org/abs/2604.21931v1\"\u003earXiv\u003c/a\u003e\u003c/td\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003ctd\u003e📄 论文\u003c/td\u003e\n\u003ctd\u003eSeeing Without Eyes: 4D Human-Scene Un…\u003c/td\u003e\n\u003ctd\u003eCV\u003c/td\u003e\n\u003ctd\u003e\u003c/td\u003e\n\u003ctd\u003e\u003ca href=\"https://arxiv.org/abs/2604.21926v1\"\u003earXiv\u003c/a\u003e\u003c/td\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003ctd\u003e📄 论文\u003c/td\u003e\n\u003ctd\u003eTemporal Taskification in Streaming Co…\u003c/td\u003e\n\u003ctd\u003eLG\u003c/td\u003e\n\u003ctd\u003e\u003c/td\u003e\n\u003ctd\u003e\u003ca href=\"https://arxiv.org/abs/2604.21930v1\"\u003earXiv\u003c/a\u003e\u003c/td\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003ctd\u003e📄 论文\u003c/td\u003e\n\u003ctd\u003eFine-Tuning Regimes Define Distinct Co…\u003c/td\u003e\n\u003ctd\u003eLG\u003c/td\u003e\n\u003ctd\u003e\u003c/td\u003e\n\u003ctd\u003e\u003ca href=\"https://arxiv.org/abs/2604.21927v1\"\u003earXiv\u003c/a\u003e\u003c/td\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003ctd\u003e📄 论文\u003c/td\u003e\n\u003ctd\u003eWhen Prompts Override Vision: Prompt-I…\u003c/td\u003e\n\u003ctd\u003eAI\u003c/td\u003e\n\u003ctd\u003e\u003c/td\u003e\n\u003ctd\u003e\u003ca href=\"https://arxiv.org/abs/2604.21911v1\"\u003earXiv\u003c/a\u003e\u003c/td\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003ctd\u003e🌟 项目\u003c/td\u003e\n\u003ctd\u003ealchaincyf/huashu-design\u003c/td\u003e\n\u003ctd\u003eGitHub\u003c/td\u003e\n\u003ctd\u003eHTML\u003c/td\u003e\n\u003ctd\u003e\u003ca href=\"https://github.com/alchaincyf/huashu-design\"\u003eGitHub\u003c/a\u003e\u003c/td\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003ctd\u003e🌟 项目\u003c/td\u003e\n\u003ctd\u003eConardLi/garden-skills\u003c/td\u003e\n\u003ctd\u003eGitHub\u003c/td\u003e\n\u003ctd\u003eJavaScript\u003c/td\u003e\n\u003ctd\u003e\u003ca href=\"https://github.com/ConardLi/garden-skills\"\u003eGitHub\u003c/a\u003e\u003c/td\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003ctd\u003e🌟 项目\u003c/td\u003e\n\u003ctd\u003ecosmicstack-labs/mercury-agent\u003c/td\u003e\n\u003ctd\u003eGitHub\u003c/td\u003e\n\u003ctd\u003eTypeScript\u003c/td\u003e\n\u003ctd\u003e\u003ca href=\"https://github.com/cosmicstack-labs/mercury-agent\"\u003eGitHub\u003c/a\u003e\u003c/td\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003ctd\u003e🔥 热帖\u003c/td\u003e\n\u003ctd\u003eI cancelled Claude: Token issues, decl…\u003c/td\u003e\n\u003ctd\u003eHN\u003c/td\u003e\n\u003ctd\u003e950 pts\u003c/td\u003e\n\u003ctd\u003e\u003ca href=\"https://news.ycombinator.com/item?id=47892019\"\u003eHN\u003c/a\u003e\u003c/td\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003ctd\u003e🔥 热帖\u003c/td\u003e\n\u003ctd\u003eGoogle plans to invest up to $40B in A…\u003c/td\u003e\n\u003ctd\u003eHN\u003c/td\u003e\n\u003ctd\u003e805 pts\u003c/td\u003e\n\u003ctd\u003e\u003ca href=\"https://news.ycombinator.com/item?id=47892074\"\u003eHN\u003c/a\u003e\u003c/td\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003ctd\u003e🔥 热帖\u003c/td\u003e\n\u003ctd\u003eThe West forgot how to make things, no…\u003c/td\u003e\n\u003ctd\u003eHN\u003c/td\u003e\n\u003ctd\u003e598 pts\u003c/td\u003e\n\u003ctd\u003e\u003ca href=\"https://news.ycombinator.com/item?id=47907879\"\u003eHN\u003c/a\u003e\u003c/td\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003ctd\u003e🔥 热帖\u003c/td\u003e\n\u003ctd\u003eNew 10 GbE USB adapters are cooler, sm…\u003c/td\u003e\n\u003ctd\u003eHN\u003c/td\u003e\n\u003ctd\u003e585 pts\u003c/td\u003e\n\u003ctd\u003e\u003ca href=\"https://news.ycombinator.com/item?id=47899053\"\u003eHN\u003c/a\u003e\u003c/td\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003ctd\u003e🔥 热帖\u003c/td\u003e\n\u003ctd\u003eSabotaging projects by overthinking, s…\u003c/td\u003e\n\u003ctd\u003eHN\u003c/td\u003e\n\u003ctd\u003e521 pts\u003c/td\u003e\n\u003ctd\u003e\u003ca href=\"https://news.ycombinator.com/item?id=47890799\"\u003eHN\u003c/a\u003e\u003c/td\u003e\n\u003c/tr\u003e\n\u003c/tbody\u003e\n\u003c/table\u003e\n",
  "summary": "📅 生成时间：2026-04-26 21:17 (Asia/Shanghai) | 数据来源：arXiv · GitHub · HackerNews · 科技媒体 · 大厂博客\n📄 一、arXiv 最新论文 1. Seeing Fast and Slow: Learning the Flow of Time in Videos 方向：arXiv/CV 摘要：How can we tell whether a video has been sped up or slowed down? How can we generate videos at different speeds? Although videos have been central to modern computer vision research, little attention has been paid to perceiving and controlling the passage of time."
}