8
0[] Welcome to Daily Tech briefing. Today we have a packed show covering Google I/O 2026 announcements, major AI talent moves, and breakthroughs in AI-powered science. Let us dive in.
欢迎收听每日科技英文速读。今天的节目将带来 Google I/O 2026 的重磅发布、AI 领域重大人才变动，以及 AI 驱动科学突破的最新进展。
---
[] OpenAI co-founder Andrej Karpathy joins Anthropic's pre-training (预训练) team. Andrej Karpathy, the AI researcher who co-founded OpenAI and previously led AI at Tesla, has joined Anthropic to work on pre-training (预训练). Karpathy announced the move on X, saying the next few years at the frontier (前沿) of large language models will be especially formative (形成性的、关键的). At Anthropic, he will build a team focused on using Claude to accelerate pre-training (预训练) research.
OpenAI 联合创始人 Andrej Karpathy 加入 Anthropic 预训练团队.OpenAI 联合创始人、前 Tesla AI 负责人 Andrej Karpathy 已加入 Anthropic，参与预训练工作。他是少数能弥合 LLM 理论与大规模训练实践之间差距的研究者之一。
---
[] Google launches Gemini 3.5 Flash, betting its next AI wave on agents, not chatbots. Google launched Gemini 3.5 Flash at Google I/O 2026, its strongest model yet for coding and autonomous (自主的) AI agents. The model can independently execute coding pipelines, manage research projects, and in internal tests, build an operating system entirely from scratch. It outperforms Google's previous frontier (前沿) model 3.1 Pro on nearly all benchmarks and is 4 times faster than other frontier (前沿) models.
Google 发布 Gemini 3.5 Flash，押注 AI 的下一个浪潮是智能体.Google 在 I/O 2026 上发布了 Gemini 3.5 Flash，这是其最强的编程和自主 AI 智能体模型。它在几乎所有基准上都超过此前的前沿模型 3.1 Pro，速度是其他前沿模型的 4 倍。
---
[] Google's Gemini Omni turns images, audio, and text into video. Google unveiled Gemini Omni at I/O 2026, a new family of multimodal (多模态的) models that can reason across text, images, audio, and video to generate consistent output. The first model, Omni Flash, can render 10 seconds of video and is rolling out to the Gemini app, YouTube Shorts, and Flow. CEO Sundar Pichai said Omni represents the next step from predicting text to simulating reality.
Google Gemini Omni：将图像、音频和文本转化为视频.Google 在 I/O 2026 上发布了 Gemini Omni 多模态模型系列，可跨文本、图像、音频和视频进行推理并生成一致的输出。首发 Omni Flash 可渲染 10 秒视频。
---
[] Google's Genie world model (世界模型) can now simulate (模拟) real streets with Street View. Google is integrating Street View data into its Project Genie world model (世界模型), allowing users to create interactive simulations of real-world locations. Announced at Google I/O 2026, the feature lets users explore environments, change weather conditions, and simulate (模拟) rare scenarios (罕见场景). Google has collected over 280 billion Street View images across 110 countries over 20 years.
Google Genie 世界模型结合 Street View 模拟真实街道.Google 将 Street View 数据整合到 Project Genie 世界模型中，用户可创建真实地点的交互式模拟。Google 已在 110 个国家收集了超过 2800 亿张 Street View 图像。
---
[] Google introduces Gemini Spark, a 24/7 agentic (智能体的) assistant with Gmail integration. Google announced Gemini Spark at I/O 2026, a new personal AI agent that runs 24/7 on dedicated virtual machines in Google Cloud. Users can email Spark through a dedicated Gmail address, and integrate with Gmail, Google Docs, and other Workspace products out of the box (开箱即用). CEO Sundar Pichai described Spark as capable of taking on long-horizon tasks with minimal oversight.
Google 发布 Gemini Spark：全天候智能体助手.Google 在 I/O 2026 上发布了 Gemini Spark，一个可在 Google Cloud 虚拟机上全天候运行的个人 AI 智能体。用户可通过 Gmail 与它交互，并开箱即用地集成 Workspace 产品。
---
[] SandboxAQ brings drug discovery (药物发现) models to Claude. SandboxAQ, an Alphabet spinout founded about five years ago, has partnered with Anthropic to integrate its scientific AI models into Claude. The company produces Large Quantitative Models that are physics-grounded, capable of running quantum chemistry (量子化学) calculations and simulating molecular dynamics (分子动力学). This makes drug discovery (药物发现) accessible to researchers without specialized computing infrastructure.
SandboxAQ 将药物发现模型接入 Claude.从 Alphabet 分拆的 SandboxAQ 与 Anthropic 合作，将其科学 AI 模型集成到 Claude 中。基于物理定律的大型定量模型可进行量子化学计算和分子动力学模拟。
---
[] OpenAI makes it easier to check AI-generated images. OpenAI announced two new measures to help detect AI-generated imagery. The company committed to the C2PA open standard, adding a clear signal in metadata (元数据) that an image was generated by AI. It is also partnering with Google to include SynthID, an invisible watermark (水印) designed to persist even when bad actors attempt to remove it. A public verification tool is in preview.
OpenAI 让 AI 图像检测更加容易.OpenAI 宣布了两项新措施检测 AI 生成图像：采用 C2PA 开放标准，并与 Google 合作引入 SynthID 隐形水印。OpenAI 还预告了公开验证工具。
---
[] Google Search as you know it is over. Google is transforming Search from a list of blue links into an AI-powered experience at I/O 2026. The new Search features conversational (对话式的) answers, autonomous (自主的) agents, and interactive interfaces. Users can create, customize, and manage AI agents directly on the platform. The new Search is powered by Gemini 3.5 Flash.
你所知道的 Google 搜索已经终结.Google 在 I/O 2026 上宣布将搜索从蓝色链接列表转变为 AI 驱动的新体验，包括对话式答案、自主智能体和交互界面，由 Gemini 3.5 Flash 驱动。
---
[] That's all for today. See you tomorrow, and thank you for listening.
以上就是今天的内容。感谢你的收听，我们明天见。
---
Vocabulary Notes:
1. pre-training 预训练
2. frontier 前沿
3. autonomous 自主的
4. multimodal 多模态的
5. world model 世界模型
6. agentic 智能体的
7. drug discovery 药物发现
8. provenance 来源、出处
9. watermark 水印
10. conversational 对话式的
11. simulate 模拟
12. pipeline 流水线、流程
13. formative 形成性的、关键的
14. benchmark 基准测试
15. avatar 数字头像
16. deepfake 深度伪造
17. rare scenarios 罕见场景
18. virtual machine 虚拟机
19. out of the box 开箱即用
20. quantum chemistry 量子化学
21. molecular dynamics 分子动力学
22. metadata 元数据
23. proactively 主动地
24. transformation 转型、变革
