infoAI 日报 2026-08-142026年08月14日105论文 5 / 官方 5 / 英文媒体 0 / 中文媒体 0本页聚合 AI 前沿信息：arXiv 论文、OpenAI/DeepMind/Meta/HuggingFace 官方博客、国内外 AI 媒体，均为最近 7 天内发布，附中文摘要与重要度评级。State of Open Models: Summer 2026 ObservationsHugging Face Blog2026-08-14T00:00:00Zhttps://huggingface.co/blog/state-of-open-models-summer-2026总结了2026年夏季开源模型的现状，涵盖最新进展和趋势。这为关注开源AI生态的读者提供了宏观视角。Record, train, and deploy from one place with Strands Agents, LeRobot, and Hugging Face Storage BucketsHugging Face Blog2026-08-13T17:16:04Zhttps://huggingface.co/blog/amazon/strands-lerobot-streaming-data-loopHugging Face 推出 Strands Agents 与 LeRobot 集成，支持从单一平台记录、训练和部署机器人智能体。该方案结合存储桶简化数据管理，旨在降低机器人开发门槛，加速具身智能应用落地。Introducing Gemini 3.7 FlashGoogle DeepMind Blog2026-08-13T17:04:18Zhttps://deepmind.google/blog/introducing-gemini-3-7-flash/Google DeepMind 发布 Gemini 3.7 Flash 模型，主打高效推理与多模态能力。该模型针对实时应用优化，提升响应速度与准确性，进一步巩固 Gemini 系列在 AI 前沿的竞争力。Test moderators use AI-generated writing to judge literacy standardsNew Scientist AI2026-08-14T14:28:46Zhttps://www.newscientist.com/article/2584776-test-moderators-use-ai-generated-writing-to-judge-literacy-standards/?utm_campaign=RSS|NSNS&utm_content=artificial-intelligence&utm_medium=RSS&utm_source=NSNS英国政府项目为降低关键阶段2评估成本，提议使用ChatGPT生成的脚本替代真实学生作文，作为评分审核员的示例。此举引发对评估公平性和AI使用边界的争议。Tech Visionary Says the Big AI Labs Don’t Get What People WantWired AI2026-08-14T15:00:00Zhttps://www.wired.com/story/tech-visionary-says-the-big-ai-labs-dont-get-what-people-want/科技思想家Tim O'Reilly认为大型AI实验室未能理解用户真正需求，他虽因AI冲击出版业而受损，但仍支持开源AI。他强调开源模式对AI发展的重要性，批评当前大厂封闭策略。These ‘Masturbation Consultants’ Were Hired to Pleasure Themselves With AIWired AI2026-08-14T10:45:00Zhttps://www.wired.com/story/these-masturbation-consultants-were-hired-to-pleasure-themselves-using-ai/Joi AI公司雇佣10人进行为期一个月的“健康”研究，要求他们使用AI伴侣自慰，声称此举有助解决“男性孤独”。该事件引发对AI伦理和商业化边界的争议。People Are ‘Marrying’ Chatbots. These Lawmakers Want to Stop ThemWired AI2026-08-14T10:15:00Zhttps://www.wired.com/story/people-are-marrying-chatbots-these-lawmakers-want-to-stop-them/美国法律目前不承认人类与AI的婚姻，部分共和党州议员正起草立法以维持现状。此举反映AI伴侣关系引发的社会和法律挑战，引发对AI情感交互的讨论。Z.ai Ships GLM-5.3 Without Retraining the Base Model: Better at Complex Coding and Long-Horizon TasksMarkTechPost2026-08-14T08:03:22Zhttps://www.marktechpost.com/2026/08/14/z-ai-ships-glm-5-3-without-retraining-the-base-model-better-at-complex-coding-and-long-horizon-tasks/Z.ai发布GLM-5.3，未重训743B基础模型，仅通过扩展后训练提升性能，在Terminal-Bench 3.0上从4.6升至28.3，DeepSWE v1.1从46.2升至66.9。网络安全能力超预期，CyberGym达84.5%，权重约两周后发布。Bring your spreadsheet data to life with Sheets canvasGoogle AI Blog2026-08-13T16:45:00Zhttps://blog.google/products-and-platforms/products/workspace/sheets-canvas-for-google-sheets-spreadsheets/Google AI博客展示Sheets canvas功能，通过视频演示将电子表格数据可视化，提升数据处理和呈现效率。How Claude’s text watermark worksAnthropic News2026-08-14T00:00:00Zhttps://www.anthropic.com/news/claude-text-watermarkAnthropic官方解释Claude文本水印的工作原理，旨在提升AI生成内容的可追溯性，可能涉及嵌入和检测机制。