x-Susan/STEM MOM 来源下的内容 - news.news

暂无通知

x-Susan/STEM MOM

Susan/STEM MOM

1周前

2 个prompt把太阳系做出来了，还带mouse motion。你们去玩吧，我要去喝两杯。 2 个prompt把太阳系做出来了，还带mouse motion。你们去玩吧，我要去喝两杯。 #太阳系 #mouse motion #喝酒

Susan/STEM MOM

1周前

Dave's Garage对于Deepseek 的评价，10个小时破百万。他还是不错的，算是我的硬核博主单。 Title: Deepseek R1 Explained by a Retired Microsoft Engineer Author: Dave's Garage Upload Date: 2025-01-28T02:47:25Z URL: Let me help break th Dave's Garage对于Deepseek 的评价，10个小时破百万。他还是不错的，算是我的硬核博主单。 Title: Deepseek R1 Explained by a Retired Microsoft Engineer Author: Dave's Garage Upload Date: 2025-01-28T02:47:25Z URL: Let me help break this detailed technical presentation into time blocks, providing context for each section to help build a comprehensive understanding: 0:00-2:30 Introduction and Context Setting Dave Plumber, a retired Microsoft engineer from the MS-DOS and Windows 95 era, introduces Deep Seek R1 as a "Sputnik moment" in AI development. He frames this Chinese open-source AI model as a significant technological milestone that's challenging Western assumptions about AI dominance. 2:30-5:00 Economic Impact and Market Significance A critical discussion of how Deep Seek R1's reported $6 million development cost has rattled the tech industry, particularly affecting Nvidia and Microsoft stock prices. The presenter draws an apt analogy: it's like building a Ferrari in your garage using Chevy parts, which challenges the entire premium AI development ecosystem. 5:00-8:30 Technical Architecture Explanation Details Deep Seek R1's fundamental architecture as a distilled language model. Dave explains how it leverages larger AI models like GPT-4 or Meta's Llama as scaffolding, using an insightful apprenticeship analogy to explain model distillation - where a smaller model learns from larger ones without needing to replicate their entire knowledge base. 8:30-12:00 Training Methodology Deep Dive Explores how Deep Seek R1 combines insights from multiple AI architectures, comparing it to assembling a panel of experts to train one exceptional student. This section includes practical demonstrations of the model's capabilities, including its handling of sensitive topics like Tiananmen Square. 12:00-15:30 Hardware Requirements and Accessibility Detailed discussion of running Deep Seek R1 on various hardware configurations, from high-end AMD Threadrippers to consumer-grade MacBooks and even $249 Ora Nano systems. This section emphasizes the model's accessibility compared to traditional AI infrastructure requirements. 15:30-19:00 Limitations and Trade-offs Thoughtful analysis of the model's potential drawbacks, including increased likelihood of hallucinations and limitations in specialized knowledge domains. Dave draws parallels to the early personal computing era, suggesting Deep Seek R1 might represent a similar democratizing force in AI. 19:00-22:30 Global Implications and Competition Examines how Deep Seek R1's release affects the global AI landscape, particularly its impact on American tech companies and their business models. Discusses the potential democratization of AI access worldwide. 22:30-25:00 Critical Analysis and Skepticism Addresses skepticism about Deep Seek's development claims, including the possibility of undisclosed state-level support and strategic implications for global AI competition. 25:00-27:00 Conclusion and Channel Information Wraps up with final thoughts on Deep Seek R1's significance and includes standard YouTube engagement requests and information about Dave's other content, including his book on autism spectrum experiences. This timeline breakdown reveals how Dave skillfully builds from basic concepts to complex implications, helping viewers understand both the technical and strategic significance of Deep Seek R1 in the evolving AI landscape. #DeepSeek #Dave's Garage #硬核博主

Susan/STEM MOM

4周前

蛋小黄是搞金融的。其实他说的一点在于：金融，房地产，互联网，传统广告媒体这几个失业主要的行业房地产其实是最容易想通的，就是“建完了”。经济下行是一部分，最主要的是经济结构正在转变。所以不应该期望去重新找到一样的工作。这点我和几个朋友还是挺认同的。 NLP会成为如office 这样的存在。NLP的实质是什么？非结构化数据转为结构化数据：分析，计算深度语义分析和生成能力，在用接近人类语 蛋小黄是搞金融的。其实他说的一点在于：金融，房地产，互联网，传统广告媒体这几个失业主要的行业房地产其实是最容易想通的，就是“建完了”。经济下行是一部分，最主要的是经济结构正在转变。所以不应该期望去重新找到一样的工作。这点我和几个朋友还是挺认同的。 NLP会成为如office 这样的存在。NLP的实质是什么？非结构化数据转为结构化数据：分析，计算深度语义分析和生成能力，在用接近人类语言的方式实现互动。打个比方，蛋小黄的金融业。绝大部份80%的从业员，不是将非结构化数据转化为结构数据，比如银行的柜台，保险的填单，包括银行后台交易中的一些审计核对。要么就是语义分析和生成，比如客户关系经理。想通这个其实就好分析了。人类工业化以来，出现科技范式转移，是不会因为人情回头的。因为这些技术你不用你的竞争对手会用。 #金融 #房地产 #经济结构转变

Susan/STEM MOM

1个月前

提高认知最便宜，最直接，最快能入手的一项“教育”不是自学，而是反思性写作Reflective Writing。我几个月前密集的说过，再继续说。你有没有发现很多知乎/微博大v越写越厉害，认知越来越高的样子。我们的语文教育真的有问题，现在要补上这一课，靠反思性写作。目前北美的至少50%的大学将此并入大学日常教纲。 提高认知最便宜，最直接，最快能入手的一项“教育”不是自学，而是反思性写作Reflective Writing。我几个月前密集的说过，再继续说。你有没有发现很多知乎/微博大v越写越厉害，认知越来越高的样子。我们的语文教育真的有问题，现在要补上这一课，靠反思性写作。目前北美的至少50%的大学将此并入大学日常教纲。 #认知提升 #反思性写作 #教育改革

Susan/STEM MOM

1个月前

昨天宝玉老师说这个先用克劳德artifacts实现，再慢慢考虑自己搭环境这件事情，我感觉很靠谱。很符合快乐学习的原理。 昨天宝玉老师说这个先用克劳德artifacts实现，再慢慢考虑自己搭环境这件事情，我感觉很靠谱。很符合快乐学习的原理。 #克劳德artifacts #自己搭环境 #快乐学习

Susan/STEM MOM

1个月前

这个问题我其实最同意初码的看法。原因是我也是在贴片厂周围长大的人，中国已经把这个做成一个呼吸一般的生态。人才，产业链，知识，所有的配套，如呼吸一样。尤其是住在深圳的人，感受最深。以前我住华强北，就我那个那么小的一个包，开车去南山盐田，就是顺路就给朋友带几个配套器件，下楼吃个饭的功夫。这种事情其实扩大到几千万人，难道不就是生态吗？软件的，硬件的，知识的。 这个问题我其实最同意初码的看法。原因是我也是在贴片厂周围长大的人，中国已经把这个做成一个呼吸一般的生态。人才，产业链，知识，所有的配套，如呼吸一样。尤其是住在深圳的人，感受最深。以前我住华强北，就我那个那么小的一个包，开车去南山盐田，就是顺路就给朋友带几个配套器件，下楼吃个饭的功夫。这种事情其实扩大到几千万人，难道不就是生态吗？软件的，硬件的，知识的。 #初码 #贴片厂 #生态

热搜榜