单词学习：![[79124dbb-cd32-480a-a968-926b52a44258.mp3]]

新闻播报：![[89841b28-70cb-45c1-ba24-6b842eaa8af1.mp3]]

AI应用新进展：有道同传用户破两千万，DeepSeek推视觉压缩技术

简报：

• 网易有道词典公布数据，旗下“AI同传”功能累计使用用户已突破2000万，该功能支持71种语言与125种口音识别，翻译精准度达98%。
• DeepSeek公司发布了全新的文档理解模型DeepSeek-OCR，并引入“视觉记忆压缩”机制，通过将长文本压缩到图像空间中，旨在解决大型语言模型处理超长文本的内存瓶颈问题。
• 该压缩技术可将一篇千字文章压缩为100个视觉Token，实现10倍压缩，同时模型解压缩后仍能恢复97%的原文。

相关链接：

• 网易有道词典AI同传用户突破2000万
• 颠覆LLM长文本处理！DeepSeek-OCR发布“视觉记忆压缩”机制，或破解AI记忆瓶颈

New Advances in AI Applications: Youdao Simultaneous Interpretation Users Exceed 20 Million, DeepSeek Introduces Visual Compression Technology

Briefing:

• NetEase Youdao Dictionary announced that its "AI Simultaneous Interpretation" feature has accumulated over 20 million users. This function supports 71 languages and 125 accent recognitions, achieving a translation accuracy of 98%. 🗣️
• DeepSeek has released a new document understanding model, DeepSeek-OCR, introducing a "Visual Memory Compression" mechanism. By compressing long texts into image space, it aims to address the memory bottleneck issues faced by large language models when processing ultra-long texts. 🧠
• This compression technology can reduce a thousand-word article into 100 visual tokens, achieving a 10x compression. Furthermore, the model can still recover 97% of the original text after decompression. ✨

Related Links:

• NetEase Youdao Dictionary's AI Simultaneous Interpretation Users Exceed 20 Million
• Revolutionizing LLM Long Text Processing! DeepSeek-OCR Releases "Visual Memory Compression" Mechanism, Potentially Cracking AI Memory Bottleneck

Vidu Q2发布，支持5分钟AI故事创作且速度提升三倍

简报：

• AI视频生成平台Vidu发布新一代Vidu Q2参考生视频功能，面向专业与半专业创作者，已全球同步上线。
• 新版本在语义理解、镜头控制、内容可控性及多主体一致性上实现提升，推理速度较Q1版本提升近三倍，并保持价格不变。
• 首次推出视频延长功能，支持创作长达五分钟的视频，旨在实现从“碎片镜头”到“完整故事”的创作跨越。
• 此次升级被业内视为AI视频创作从“辅助生成”向“全流程创作”阶段加速迈进的标志。

相关链接：

• Pollo AI Has Launched Vidu Q2 and Kling 2.5 Turbo AI Video Models, Empowering Creators with Studio-Level Cinematics

Vidu Q2 Released, Supports 5-Minute AI Story Creation with Triple Speed Boost 🚀

Brief:

• AI video generation platform Vidu has launched its next-generation Vidu Q2 referential video feature, now globally available for professional and semi-professional creators.
• The new version delivers enhanced semantic understanding, camera control, content controllability, and multi-character consistency. Inference speed has increased nearly threefold compared to the Q1 version, with no change in pricing. ✨
• A new video extension feature has been introduced for the first time, supporting the creation of videos up to five minutes long. This aims to bridge the gap from "fragmented shots" to "complete stories" in creative production. 🎬
• This upgrade is seen by the industry as a significant milestone, accelerating AI video creation from an "assisted generation" phase to a "full-workflow creation" stage.

Related Link:

• Pollo AI Has Launched Vidu Q2 and Kling 2.5 Turbo AI Video Models, Empowering Creators with Studio-Level Cinematics

Adobe推出AI Foundry，为企业提供定制化AI模型并附法律保障

简报：

• Adobe宣布推出名为“AI Foundry”的企业服务，该服务基于其采用授权数据训练的Firefly模型，允许企业利用自身品牌资产和知识产权开发专属的文本、图像、视频及3D内容的生成式AI模型，并提供法律保障以降低应用风险，华特迪士尼幻想工程或成首批客户。

相关链接：

• Adobe重磅推出“AI Foundry”:企业定制生成式AI模型新时代，迪士尼或成首批客户

Adobe Launches AI Foundry, Offering Customized AI Models for Enterprises with Legal Indemnification

Brief:

• Adobe has announced "AI Foundry," an enterprise service built upon its Firefly models, trained with licensed data. This service enables businesses to leverage their brand assets and intellectual property to develop proprietary generative AI models for text, images, video, and 3D content. It also offers legal indemnification to mitigate application risks, with Walt Disney Imagineering potentially among the first clients. 🚀✨

Related Link:

• Adobe's "AI Foundry" Ushers in a New Era of Enterprise-Customized Generative AI Models, Disney Potentially Among First Clients 🔗

Google Veo 3.1推出精确编辑功能，可无缝增删视频元素

简报：

• Google即将推出Veo 3.1，其核心升级是“精确编辑”功能，允许用户在保留视频原始性的同时，添加或删除视频中的元素。
• 新功能包括智能插入元素的“Insert”工具和无缝擦除并重建背景的“Remove”工具，AI会自动处理光影、反射等复杂细节，确保编辑效果逼真。
• Veo 3.1支持高达1080p分辨率和24帧每秒的视频处理，标志着AI视频技术正从内容生成向专业后期制作转型。
• 该功能已在Flow平台上线，并将通过Gemini API、Vertex AI和Gemini应用向开发者及企业用户开放。

Google Veo 3.1 Launches Precise Editing Features for Seamless Addition and Removal of Video Elements

Briefing:

• Google is set to launch Veo 3.1, with its core upgrade being "Precise Editing" functionality, allowing users to add or remove elements within videos while preserving their original integrity. 🎬
• New features include an "Insert" tool for intelligently adding elements and a "Remove" tool for seamlessly erasing and reconstructing backgrounds. AI automatically handles complex details like lighting, shadows, and reflections to ensure realistic editing results. ✨
• Veo 3.1 supports video processing up to 1080p resolution and 24 frames per second, signaling a shift in AI video technology from content generation towards professional post-production. 🚀
• This functionality is already live on the Flow platform and will be made available to developers and enterprise users via the Gemini API, Vertex AI, and Gemini applications.

华工科技掌握硅光全栈自研能力，并推出1.6T光模块方案

简报：

• 华工科技在互动平台称，公司具备从硅光芯片到模块的全自研设计能力。
• 公司已成功推出用于1.6T光模块的单波200G自研硅光芯片，并提供多种1.6T光模块产品方案。
• 在产能方面，公司已锁定并提升海外Fab产能，同时增加海外第二备份Fab，并正加快验证国内Fab。
• 公司已发起设立上游光芯片厂商云岭光电，作为其在产业链上游的战略布局。

相关链接：

• 原文链接

HGTECH Masters Full-Stack Silicon Photonics R&D and Launches 1.6T Optical Module Solutions

Briefing:

• HGTECH stated on its interactive platform that the company possesses full in-house R&D and design capabilities, from silicon photonics chips to modules. ✨
• The company has successfully launched its self-developed single-wave 200G silicon photonics chips for 1.6T optical modules and offers multiple 1.6T optical module product solutions. 🚀
• In terms of production capacity, HGTECH has secured and increased overseas Fab capacity, while also adding a second backup overseas Fab. The company is also accelerating the verification of domestic Fabs.
• HGTECH has initiated the establishment of Yunling Optoelectronics, an upstream optical chip manufacturer, as a strategic move in the upstream segment of its industry chain. 💡

Related Links:

• Original Link

苹果发布M5芯片，全面升级iPad、Mac及Vision Pro

简报：

• 苹果公司发布M5芯片，并推出搭载该芯片的全新iPad Pro、MacBook Pro及Vision Pro系列产品，宣称是自M1芯片问世以来最大的性能飞跃。
• 新款iPad Pro性能比上一代快3.5倍，起售价约合新台币32,000元。
• 14英寸MacBook Pro强调拥有长达24小时的续航能力，起售价约合新台币51,000元。
• Vision Pro也升级至M5芯片，并搭载visionOS 26系统，起售价约合新台币119,900元。
• 所有新品已开放预购，将于10月22日起全球发售。

相关链接：

• AI效能炸裂！蘋果M5系列性能翻倍、價格更驚人

Apple Unveils M5 Chip, Revamping iPad, Mac, and Vision Pro Lines

Briefing:

• Apple Inc. has announced the M5 chip and introduced new iPad Pro, MacBook Pro, and Vision Pro series products powered by it, claiming the biggest performance leap since the M1 chip's debut. 🚀
• The new iPad Pro boasts 3.5 times faster performance than its predecessor, with a starting price of approximately NT$32,000. 💨
• The 14-inch MacBook Pro emphasizes an impressive 24-hour battery life, starting at approximately NT$51,000.
• Vision Pro has also been upgraded with the M5 chip and features visionOS 26, with a starting price of approximately NTdJyH6kDkfC5RlM16.answerText19,900.
• All new products are now available for pre-order and will be globally released starting October 22nd. 🗓️

Related Link:

• AI Performance Explodes! Apple's M5 Series Doubles Performance, Prices Even More Astonishing

OpenAI为Sora 2推出新政，赋予艺术家声像使用选择权

简报：

• OpenAI于10月20日宣布，与美国演员工会（SAG-AFTRA）及多家艺能公司合作，推出旨在保护艺术家声音和肖像权的新政策。
• 针对其声音与形象模拟技术Sora 2，新政策赋予所有艺术家、表演者及个人选择权，他们可以自由决定是否允许自己的声音和形象被AI模拟。
• 此举旨在防止未授权的复制，并回应行业对AI深伪内容日益增加的担忧。

相关链接：

• OpenAI强化Sora 2保护政策，确保艺人声音与肖像权不被侵犯

OpenAI Introduces New Policies for Sora 2, Granting Artists Voice and Likeness Usage Choice 🎤

Briefing:

• On October 20, OpenAI announced new policies, in collaboration with the Screen Actors Guild – American Federation of Television and Radio Artists (SAG-AFTRA) and several talent agencies, aimed at protecting artists' voice and likeness rights.
• Specifically for its voice and image simulation technology, Sora 2, the new policy grants all artists, performers, and individuals the option to freely decide whether their voices and likenesses can be simulated by AI.
• This move aims to prevent unauthorized replication and responds to growing industry concerns over AI deepfake content. 🛡️

Related Links:

• OpenAI Strengthens Sora 2 Protection Policies to Ensure Artists' Voice and Likeness Rights Are Not Infringed ✨

IBM 联手 Groq，以 LPU 技术为 watsonx 平台 AI 推理加速

简报：

• IBM 宣布与芯片初创公司 Groq 达成战略合作，将其语言处理单元（LPU）技术集成至 watsonx 平台。
• Groq 的 LPU 架构声称在特定 AI 推理任务上可比传统 GPU 系统快五倍以上，同时能显著降低能耗与成本。
• 此次合作将优先应用于医疗保健和零售场景，IBM 客户已可通过 Watsonx Orchestrate 访问 GroqCloud 服务，该服务也将支持 IBM 自研的 Granite 模型。

相关链接：

• IBM 携手 Groq:用 LPU 技术为企业 AI 提速五倍，全面升级 watsonx 平台

IBM Partners with Groq to Accelerate AI Inference on watsonx Platform with LPU Technology

Brief:

• IBM announced a strategic partnership with chip startup Groq to integrate its Language Processing Unit (LPU) technology into the watsonx platform. 🚀
• Groq's LPU architecture claims to be over five times faster than traditional GPU systems for specific AI inference tasks, while significantly reducing power consumption and cost. 💡
• This collaboration will prioritize applications in healthcare and retail scenarios. IBM clients can already access GroqCloud services via Watsonx Orchestrate, which will also support IBM's proprietary Granite models. 🏥

Related Link:

• IBM Joins Forces with Groq: Accelerating Enterprise AI Fivefold with LPU Technology, Fully Upgrading the watsonx Platform