介紹完了 100% 台灣技術以犀動智能打造的 Vocol AI 的語言辨識技術，就這麼剛好迎來了今年 Computex 最受矚目的 NVIDIA 主題演講。而既然這個自動逐字稿＋摘要重點其背後也有用到 NVIDIA 的硬體。電腦王阿達團隊也想到，那麼何不為直播上哀嚎沒有 AI 幫忙翻出字幕的朋友一個小忙，透過 Vocol AI 的人工智慧之力幫忙進行摘要甚至是逐字稿的工作（其實上傳完影片只花了大概 10 分鐘就搞定這篇 – 小小調整版面與用語的時間都超過 10 分鐘了…）。也算這次主題演講重點的「人工智慧之力」的簡單實例驗證。繼續閱讀 NVIDIA Computex 2023 主題演講 AI 重點摘要 – 以 Vocol AI 逐字稿技術生成報導內文。
▲圖片來源：NVIDIA
NVIDIA Computex 2023 主題演講 AI 重點摘要 – 以 Vocol AI 逐字稿技術生成
https://www.youtube.com/live/QSWzSRnGEFo?feature=share
NVIDIA 首席執行官 Jensen Huang 在 Computex 上發表主旨演講，強調該公司對於加速計算和人工智慧的重點。Huang討論了 CPU 規模化的終結以及加速計算的重要性，以及該公司的軟體定義資料中心和生命週期最佳化。他展示了正在全面生產的 H100 電腦，以及專為大規模人工智慧應用而設計的 Grace Hopper 超級計算機。
Huang 還展示了生成式人工智慧的突破，包括文字轉音樂和文字轉影片的功能，並討論了人工智慧對技術以外行業的影響。NVIDIA 通過 Grace Hopper 超級計算機和 NVIDIA Maxine 3D 將生成式人工智慧引入全球每個資料中心，後者使用雲服務將 2D 影片轉換為 3D。該公司還與 SoftBank 合作，在 SoftBank 的全球資料中心中實施生成式人工智慧和軟體定義的 5G 堆棧。
以下內容均使用 Vocal AI 節錄重點並翻譯為中文，完整英文逐字稿請點我（使用 Vocal AI 辨識）：
《遊戲與加速計算的未來：一個臨界點》
NVIDIA 首席執行官 Jensen Huang 談到了遊戲的未來、IBM System 360 以及當今計算機行業的兩個基本趨勢。他解釋了 CPU 規模化的終結和加速計算的前進道路，以及資料中心即為計算機，使得加速計算成為一個完整的堆棧問題。Huang 還強調了保護軟體投資的重要性和資料中心方程式的複雜性。最後，他展示了大規模語言模型的加速計算的成本效益，並強調購買得越多，節省的成本越多。
00:00:00->00:16:11
《NVIDIA 的加速計算：革新 AI 和行業的未來》
NVIDIA 首席執行官 Jensen Huang 在 Computex 上發表主旨演講，強調該公司致力於加速計算和人工智慧。Huang 討論了該公司的軟體定義資料中心和生命週期優化，以及不同行業對 NVIDIA GPU 的高需求。他還展示了正在全面生產的H100 電腦，並討論了該公司致力於推進計算性能的重要性。Huang 還談到了生成式人工智慧的突破，以及在不同模態之間轉換信息的能力。他展示了文字轉音樂和文字轉影片的 AI 功能的示例。
00:16:12->00:36:00
《NVIDIA 推出具有 600GB 內存的 Grace Hopper AI 超級計算機》
NVIDIA 首席執行官 Jensen Huang 討論了生成式人工智慧的重要性以及該公司的新款 Grace Hopper 超級計算機，該計算機具有 600GB 的內存，專為大規模人工智慧應用而設計。該計算機可以擴展到一個艾克斯佛洛普轉換引擎，提供144TB 的內存。Huang 表示，Google Cloud、Meta 和 Microsoft 將是第一批獲得該計算機訪問權限的公司，該計算機將用於開拓性的人工智慧研究。Huang 還討論了人工智慧對技術以外行業的影響，例如數字生物學，以及新計算時代的易用性。
00:36:01->00:53:44
《NVIDIA 的 Grace Hopper 和資料中心中生成式 AI 的未來》
NVIDIA通過其新技術革新了我們在連接、溝通和協作方面的方式。該公司正在通過其 Grace Hopper 超級計算機和 NVIDIA Maxine 3D 將生成式 AI 引入全球每個資料中心。Maxine 3D 使用雲服務將 2D 影片轉換為 3D，從而在視像會議中創造出增強的深度和存在感。此外，NVIDIA 正在與 SoftBank 合作，在 SoftBank 的全球資料中心中實施生成式 AI 和軟體定義的 5G 堆棧。NVIDIA 還宣布了 NVIDIA MGX，一種用於加速計算的開放式模塊化服務器設計規範。該公司使用其 Spectrum 4 交換機推出了一種新型以太網，以在每個資料中心中實現 AI 工作負載。
00:53:45->01:10:47
《NVIDIA 在高性能計算、企業 AI 和 Omniverse 方面的進展》
NVIDIA 宣布了一個新的以太網系統，旨在將高性能計算的能力帶入以太網市場。該系統由 Spectrum 4 交換機和Bluefield 3 SmartNIC 或 DPU 組成，共同提供自適應路由和擁塞控制功能。NVIDIA 還通過其雲服務 NVIDIA AI Foundation 幫助行業構建定制語言模型，並將其企業級和企業安全軟體堆棧 NVIDIA AI Enterprise 集成到主要雲提供商中。該公司還在開發 Omniverse，這是一個數字孿生平台，允許用戶在虛擬環境中構建、設計和操作工廠。
01:10:47->01:29:36
《NVIDIA 與 WPP 合作構建基於 AI 的內容生成引擎》
NVIDIA 與全球最大的廣告代理公司 WPP 合作，利用 Omniverse 和生成式AI構建內容生成引擎。該引擎集成了 Adobe Firefly、Getty 和 Shutterstock 等合作夥伴的工具，為廣告應用生成獨特的內容，以滿足不同用戶的需求。NVIDIA 還在製造業中使用 Omniverse 和 AI，電子製造商 Pegatron 使用 NVIDIA AI 和 Omniverse 將其工廠數字化。NVIDIA 還宣布了他們的機器人平台 NVIDIA iZak AMR，作為任何想要構建機器人的參考設計。該平台包括一款名為 Orin 的芯片、一個計算機以及 NVIDIA Nova Orin，這是一個參考系統，也是 AMR 的藍圖。整個堆棧都已構建完成，它使製造商能夠超高速度啟動工廠，持續優化操作，最大限度地提高生產線產能，同時降低成本。
01:29:36->01:45:37
英文版本
NVIDIA CEO Jensen Huang delivered a keynote speech at Computex, highlighting the company’s focus on accelerating computing and artificial intelligence. Huang discussed the end of CPU scaling and the importance of accelerated computing, as well as the company’s software-defined data center and life cycle optimization. He showcased the H100 computer, which is in full production, and the Grace Hopper supercomputer, designed for large-scale AI applications. Huang also demonstrated generative AI breakthroughs, including text-to-music and text-to-video capabilities, and discussed the impact of AI on industries beyond technology. NVIDIA is introducing generative AI to every data center in the world through its Grace Hopper supercomputer and NVIDIA Maxine 3D, which converts 2D videos to 3D using cloud services. The company is also partnering with SoftBank to implement generative AI and software-defined 5G stack into SoftBank’s data centers worldwide.
The Future of Video Games and Accelerated Computing: A Tipping Point
NVIDIA’s CEO, Jensen Huang, spoke about the future of video games, the IBM System 360, and the two fundamental trends in the computer industry today. He explained that CPU scaling has ended and accelerated computing is the path forward, and that the data center is the computer, which makes accelerated computing a full stack problem. Huang also highlighted the importance of preserving software investment and the complexity of the data center equation. Finally, he demonstrated the cost-effectiveness of accelerated computing for large language models and emphasized that the more you buy, the more you save.
00:00:00->00:16:11
NVIDIA’s Accelerated Computing: Revolutionizing the Future of AI and Industry
NVIDIA CEO Jensen Huang gave a keynote speech at Computex, highlighting the company’s dedication to accelerating computing and artificial intelligence. Huang discussed the company’s software-defined data center and life cycle optimization, as well as the high demand for NVIDIA GPUs across industries. He also showcased the H100 computer, which is in full production, and discussed the company’s dedication to pushing the performance of computing forward. Huang also spoke about the breakthroughs in generative AI and the ability to transform information across modalities. He demonstrated examples of text-to-music and text-to-video AI capabilities.
00:16:12->00:36:00
NVIDIA Unveils Grace Hopper AI Supercomputer with 600GB Memory
Nvidia CEO Jensen Huang discussed the importance of generative AI and the company’s new Grace Hopper supercomputer, which has 600GB of memory and is designed for large-scale AI applications. The computer can be scaled up to an exaflop transformer engine, which would offer 144TB of memory. Huang said Google Cloud, Meta, and Microsoft will be the first companies to have access to the computer, which will be used for pioneering AI research. Huang also discussed the impact of AI on industries beyond technology, such as digital biology, and the ease of use of the new computing era.
00:36:01->00:53:44
NVIDIA’s Grace Hopper and the Future of Generative AI in Data Centers
NVIDIA is revolutionizing the way we connect, communicate, and collaborate with its new technology. The company is introducing generative AI to every data center in the world by using its Grace Hopper supercomputer and NVIDIA Maxine 3D. Maxine 3D converts 2D videos to 3D using cloud services, creating an enhanced sense of depth and presence in video conferencing. Additionally, NVIDIA is partnering with SoftBank to implement generative AI and software-defined 5G stack into SoftBank’s data centers worldwide. NVIDIA is also announcing the NVIDIA MGX, an open modular server design specification designed for accelerated computing. The company is introducing a new type of Ethernet using its Spectrum 4 switch to enable AI workloads in every data center.
00:53:45->01:10:47
NVIDIA’s Advances in High Performance Computing, Enterprise AI, and Omniverse
NVIDIA has announced a new Ethernet system for AI, designed to bring the capabilities of high-performance computing into the Ethernet market. The system consists of the Spectrum 4 switch and the Bluefield 3 SmartNIC or DPU, which together offer adaptive routing and congestion control. NVIDIA is also helping industries build custom language models with its cloud service, NVIDIA AI Foundation, and has integrated its enterprise-grade and enterprise-secure software stack, NVIDIA AI Enterprise, into major cloud providers. The company is also developing Omniverse, a digital twin platform that allows users to build, design, and operate factories in a virtual environment.
01:10:47->01:29:36
NVIDIA and WPP Partner to Build AI-Enabled Content Generation Engine
NVIDIA has partnered with WPP, the world’s largest advertising agency, to build a content generation engine using Omniverse and generative AI. The engine integrates tools from partners such as Adobe Firefly, Getty, and Shutterstock to generate unique content for different users for ad applications. NVIDIA is also using Omniverse and AI for manufacturing, with electronics manufacturer Pegatron using NVIDIA AI and Omniverse to digitalize their factories. NVIDIA has also announced their robot platform, NVIDIA iZak AMR, as a reference design for anyone who wants to build robots. The platform includes a chip called Orin, a computer, and the NVIDIA Nova Orin, which is a reference system, a blueprint for AMRs. The entire stack has been built, and it allows manufacturers to super accelerate factory bring-up, continuously optimize operations, maximize production line throughput, all while reducing costs.
01:29:36->01:45:37
▲圖片來源：NVIDIA
不得不說，雖然在這篇沒辦法展示給大家，可以直接搭配錄音／影片直接聆聽每個演講段落的使用體驗。但大致上應該也算火力展示了目前 AI 可以快速整理會議／發表會的原始實力 – 當然，這裡面肯定有一些錯誤，以及漏掉的地方（大家可以參考下一頁會有逐字稿，另外，標注的時間標籤會以演講開始時為主，所以可能會有一些時間誤差）。
不過對於需要快速掌握資訊的人而言，Vocol AI 所提供的超高整理效率與歸納的能耐，的確是非常值得試用的好服務。關於 NVIDIA Computex 2023 主題演講完整內容，也可以參考官方的新聞部落格。
延伸閱讀：
自動逐字稿＋GPT 摘要重點！Vocol.ai 最強 AI 會議整理神器使用分享教學（免費點數超夠用快來搶）
