让变化经过筛选，
再抵达你。

博科尼大学研究者与 OpenAI Economic Research 对超过 1,000 名一年级学生开展随机实验。官方摘要称，ChatGPT 访问提高了作业质量与连贯性，因果推理训练则带来更多独特想法，两者结合时效果并存。研究使用特定商业案例和评分方法，结论不应自动推广到所有课程。

Researchers at Bocconi University, working with OpenAI Economic Research, ran a randomized experiment with more than 1,000 first-year students. The published summary reports that ChatGPT access improved quality and coherence, while causal-reasoning training produced more distinctive ideas, with both effects appearing together. The study used one business case and evaluation design, so its findings should not be generalized automatically to every course.

Google 将航班价格提醒直接加入 Search 的 AI Mode，并新增积分或里程价格查看与酒店预订入口。用户可在对话中描述时间与目的地，再从 300 多家航空公司和旅行网站的合作数据中比较选项；价格与库存仍会实时变化，最终信息应以预订页面为准。

Google is bringing flight-price alerts into AI Mode in Search, alongside points-or-miles views and hotel-booking entry points. Users can describe dates and destinations conversationally and compare options sourced from more than 300 airline and travel partners. Prices and availability remain dynamic, so final details should be confirmed on the booking page.

Google Labs 发布 Expert Intelligence，让用户在 Gemini Notebook 中调用合作作者、出版物与领域专家的内容，包括用户已购买的部分书籍。该功能强调来源与授权关系，但“可信”仍不等同于无误；阅读者仍需区分原文、摘要和模型生成的解释。

Google Labs has introduced Expert Intelligence, which lets Gemini Notebook draw from participating authors, publications, and domain experts, including selected titles a user has purchased. The feature emphasizes source and licensing relationships, but trusted does not mean infallible; readers should still distinguish original material, summaries, and model-generated interpretation.

Google 为 Flow 的视频编辑工作流加入首尾帧控制、1080p 与 4K 导出，以及低额度 360p 草稿后放大到 720p 的流程。官方称这些更新基于 Gemini Omni 1.1 Flash，可帮助维持人物与叙事连续性；实际一致性与质量仍取决于素材、提示和生成结果。

Google has added start-and-end frame control, 1080p and 4K export, and a lower-credit 360p drafting flow that can upscale selected clips to 720p in Flow. Google says the Gemini Omni 1.1 Flash updates help maintain character and narrative continuity; actual consistency and quality still depend on source material, prompts, and generated output.

AWS 发布一套由 Amazon Quick、Skills、MCP 与 fal 组成的创意工作流方案，示例涵盖八格分镜和音乐视频概念。设计重点不是单次生成速度，而是保留上下文、支持长任务，并在关键节点加入人工批准。这是一篇技术方案演示，不是生产效果的独立评测。

AWS has published a creative-workflow pattern combining Amazon Quick, Skills, MCP, and fal, demonstrated through an eight-panel storyboard and a music-video concept. The design focuses on preserving context, supporting long-running jobs, and inserting human approval at key gates rather than merely accelerating one generation step. It is a technical demonstration, not an independent production benchmark.

Google Research 发布实验性的 Planetary Prediction Engine，可从自然语言问题开始，自动完成地理数据发现、特征工程、模型训练、评估与报告。团队报告其在多项公开任务上改善基线，并在一项疾病传播预测中取得 83.3% 的 Recall@10；PPE 仍是早期研究系统，数字来自研究团队测试，不能替代领域专家和正式决策流程。

Google Research has introduced the experimental Planetary Prediction Engine, which moves from a natural-language question through geospatial data discovery, feature engineering, model training, evaluation, and reporting. The team reports improvements across public tasks and 83.3% Recall@10 in one disease-spread forecast. PPE remains early-stage research; the figures are team-reported and the system does not replace domain experts or formal decision processes.