大学

香港大学孙教授的一段话：

大学，就应该是早起吃点早餐；跑跑步；专业课认真听；公共课看看自己喜欢的杂志；中午小睡一会儿；下午参加个社团活动或打打篮球；晚上陪着喜欢的人散散步；或去自习室安安静静地看看书……

社会不需要学霸，也不认什么学生会主席，更不希望看到学生放弃学业去创业。你只要能平稳完整滴读完大学，寻找到自己所爱的人和兴趣，多去没有目的的看些能丰富自己思想的书，认识几个好得不成样子的朋友，锻炼或是塑造自己的身体，学精自己想要从事事业的专业知识……

做到这些，平淡地度过大学这几年你就已经足够优秀了。

深以为然。

Converged Infrastructures (CI) vs. Hyperconverged Infrastructures (HCI)

• Converged infrastructure relies on hardware and employs building blocks. Hyperconverged infrastructure relies on software-defined. 
• The main difference between CI and HCI involves the rack system. A large, rack-scale platform that merges compute, storage, and networking into a turnkey product is CI, while HCI normally consists of a 1U or 2U (rack-unit) systems that consolidate one or more multi-core servers with a local storage array. 
• The architecture varies between the two approaches. “Converged architecture storage is attached directly to the physical server” and "the HCI architecture has a storage controller function that runs as a service on every node in the cluster."
• HCI shares storage to all compute and virtual machines (VMs), whereas CI does not. 
• HCI is often deployed on commodity components; providing a simplified scale out architecture with commodity servers. 
• HCI is more flexible, manoeuvrable, and scalable than CI. 
• HCI does not replace CI. They contain unique benefits. 

What is Serverless Application?

A serverless application is an application implemented using serverless services, and at least five traits are characterised for such application.

• No management of server hosts or server processes are required
• Auto-scaling and auto-provisioning based on load
• Cost is based on usage
• Performance capabilities are defined in different terms other than host size or count
• Has implicit high availability

Serverless is hostless; there is no server to work with. The good thing is there is less operational load. No server to upgrade; no security patch to apply. On the other hand, different kinds of metrics need to be monitored in an application.
Serverless is stateless. When state is not stored in an application, horizontal scaling is very easy and you just spin up more instance. Being stateless also means that room for error is greatly reduced. Being stateless also means that techniques that require state can’t be used in application development; for example, it’s not possible to use HTTP sessions.
A serverless architecture where many components are integrated via the network is distributed by default. Persistence is achieved in Backend as a Service (BaaS), code is run in multiple functions, other services are used for authentications and queues, etc. Being distributed also bring high availability to the architecture.

捋捋Microsoft Build 2019

微软Build各种新闻，捋捋：比如自带Linux kernel；新的Terminal（说实话，这个等了至少20年才像个Terminal的样子）；React Native Windows；还有把IE的功能做回基于Chromium的Edge，这个比较妖艳，就好比在触摸屏上实现鼠标功能。估摸这原因很有可能是当年某些政府工程外包中的码农只上过基于IE的Web开发培训班…… 

最后，最重要的，微软也吹云Azure，伴着Office和AI一起。感觉和Google和Amazon比起来，有那么一点点村。不过，要把话说回来，在.Net和Code上是非常走心，再加上牛逼的Github。与其他几家比起来，对开发社区和人员的关怀度满满地溢出。感觉微软把Windows定位改了以后确实腾出了很多发展空间，可见重新定位对于微软来说，是防守反击的最重要，盘活了大局。

看不懂的是，把AI往云上吹。这是相信了现在的算力，还是满意了现在架构的网力呢。云AI怎么落地呢？ 不过微软最牛还是销售，各大公司CIO都是忠实的信徒。还有的可看。

过不了一会儿就是Google I/O，且看劈柴怎么批。

A Few of Facts about Eric Yuan, ZOOM

Zoom, the video conferencing company, had an IPO on Thursday April 18 2019. Shares jumped 72% on its opening day, putting the company's valuation at nearly $16 billion. Eric Yuan, Zoom's founder and CEO, owns 20.5% of his company's stock. At the price Zoom closed at Thursday, Yuan is now worth more than $3 billion.

Unlike most Silicon Valley tech unicorns, Zoom is profitable AND it has been growing like crazy. The company saw $7.5 million in profits in 2018. The company brought in $330.5 million in revenue in fiscal 2018, up 118% from $151 million in fiscal 2017.

• He moved to the United States in 1997 after his visa application was rejected 8 times. 
• He spoke very little English when he moved to the U.S. Instead of taking English lessons or going to an US graduate school, he chose to learn English from his colleagues.
  
• He founded Zoom in 2011 when he was 41 years old.
  
• He started a company in a super crowded market against incumbents such as Cisco/WebEx, Google Hangout, Microsoft Skype, and GoToMeeting, but he still succeeded.
  
• He was a key engineer and later VP of Engineer at Cisco/WebEx. According to a Forbe Article: “ Yuan was not happy with the way Cisco was managing WebEx when he left in 2011. As he said, ‘I was paid very well as a VP at Cisco. But WebEx was my baby. In 2010 and 2011, I did not see happy customers. I was very embarrassed that I spent so much time on the technology. Why are the customers not happy?’”
  
• He has 99% approval rate by his employees on Glassdoor, and won Glassdoor’s topCEO award.

AI is Not a Race

TechRepublic's article "How China tried and failed to win the AI race" in the early of this month describes a situation where both powerful countries are now entering into a AI war. 

I agree with most of the stances in terms of national AI development plans/initiatives, AI associated chips, AI-driven research efforts, workforces, fundings, and regulations. But it is not really a race at all. China and US are standing in the different stage of the technologies of Artificial Intelligence. They kicked off from the different starting points. US accumulated very fundamental research works since the born of AI. The developments of these researches have been very solid though it has been frozen to grow for couples of years. China instead woke up to join and contributed to the development of AI researches in the late years. It is unrealistic and harsh to compare both nations together. 

The article is right about interdependency. It should be great to see US and China each advances in certain areas of this technology. Competition and collaboration should come together to push the edge of AI. We should keep the door open to mutual interdependencies for the goods of AI deployments, as well as be cautious to set up ethic rules to make AI-human interactions right. 

AI should not become a weapon to drag two nations into a war. 

The Future of Cryptanalysis

It's been said that the advent of modern computing has spelled the death of the field of cryptanalysis; but the practice is still alive and well -- it's the methodology that's changed as technology has transformed the landscape. As quantum computing continues to develop, there're concerns that modern encryption could be at risk of being broken. This is because most modern encryption algorithms are based on large prime number factorization being computationally difficult, something that can be significantly sped up by quantum computing. Because of this, quantum computing would allow for significantly faster factorization and brute-force attacks on encryption keys, making the future of modern cryptography questionable in the looming quantum computing era. 

微服务 (Microservices) 和容器 (Container)

什么是微服务？

微服务是将应用程序拆分为多个服务的一种架构类型，这些服务具备构成整个应用程序的细粒度功能。每个微服务将具备针对您的应用程序的不同逻辑功能。与应用程序的所有组件和功能都在单个实例中的单体架构相比，微服务是应用程序架构领域一种更为现代的方法。您可以参考下图中单体架构与微服务架构的比较情况。

什么是容器？

微服务放置在哪里？在容器中。容器是存放软件的包，里面包含运行软件所需的一切内容，比如代码、依赖关系、库、二进制文件等等。Docker 是一种构建和运行容器的流行工具，但是 Kubernetes正快速成为行业标准，用于编排企业环境中的多个容器。与虚拟机相比，容器可以共享操作系统内核，而不是像在一个主机上构建多个虚拟机那样拥有完整的副本。虽然可以将微服务放置在多个虚拟机中，但在这种情况下通常会使用容器，因为容器占用的空间更少，启动速度也更快。

使用微服务架构的好处

微服务架构是为解决人们在单体应用程序中遇到的问题而创造的。微服务已被广泛使用，一些大型网站已将他们的单体应用程序转变为微服务。使用微服务架构的一些好处是：

• 与单体应用程序中的大型代码库相比，开发人员只需处理小型代码库。 当应用程序组件松散耦合时，开发人员可以轻松理解源代码，而不会减慢开发速度。如果使用的代码行数更少，您的 IDE 的速度显然会更快。开发人员无需处理各种功能的复杂性和依赖关系，这种情况只会在单体应用程序中出现。
• 开发人员的职责将会更加明确。 可以按照应用程序的组件或微服务来分配团队工作。代码复查速度将会加快。与单体应用程序相比，更新速度将会加快，而且无需构建和部署一切内容。
• 应用程序的技术堆栈可以通过微服务有所不同。应用程序不再需要依赖一种语言或库。只要开发人员认为合适，微服务就可以利用多种不同编程语言。可以使用如下图所示的多语言微服务。
• 持续交付将变得更加容易。 对于简单变更，使用微服务就无需像单体应用程序那样再次重新部署一切。您可以选择仅重新构建和部署需要更新的微服务。频繁更新的速度将会加快。
• 可扩展性与每个微服务无关。您可以选择根据应用程序所需的资源扩展它的每个组件。无需像单体应用程序那样为一切内容构建多个实例。扩展微服务将会有效利用可用资源，而不是像在单体应用程序中那样拥有整个应用程序的多个副本。
• 数据可以分散化处理。您可以选择为微服务使用不同的数据库/存储器。如果比起关系数据库，您的微服务更适合使用非关系型数据库，那么就可以选择这种数据库。微服务也可能只需要简单的密钥存储数据库，比如 Redis。如下图所示，您可以选择组合使用 Cloudant、MySQL 和 MongoDB。您可以利用不同的数据库来存储不同的数据类型。
• 隔离故障。 一个微服务中的错误或缺陷不会使整个系统宕机。如果采用松散耦合的组件，您的应用程序中的微服务出现错误时，其他微服务不太可能受到影响，因为它们都在自己的容器中，不会完全依赖彼此。而对于单体应用程序，如果没有正确找出缺陷或错误，就会使整个应用程序流程宕机。

使用微服务架构的弊端

在使用微服务解决单体架构的一些问题时，每种微服务都存在一系列问题。如果您试图将单体应用程序拆分为微服务，那么第一个挑战就是如何拆分。您可以选择将它们拆分为多个业务功能，比如一个微服务处理批次，另一个微服务处理支付服务。最后，您的组件应该只具有一小部分的功能或责任。

在微服务架构中看到的一些问题如下：

• 一旦微服务数量增长，就会难以进行跟踪。持续集成和持续交付的初始设置工作也并非易事，因为您需要处理拥有多个微服务所带来的额外复杂性。
• 复杂性。微服务需要加强协作，尤其是在有多个团队参与的时候。如果需要与其他微服务交互，那么微服务还会引进更多的网络调用，而在单体应用程序中则不会出现这种情况。部署微服务并不像部署应用程序的一个实例那样简单。您还需要考虑其他很多问题：如何处理各个微服务之间的通信，解决错误以避免中断其他微服务，以及在每个组件中添加更多测试用例。
• 找到并跟踪应用程序中的缺陷/错误。如果您的微服务只有一条路径，那么查找起来会比较容易，但如果一个微服务与其他多个微服务进行通信，仅查找错误就会耗费大量时间。
• 进行微服务路由需要完成更多工作。您需要花时间来配置和控制微服务的流动。您还需要持续跟踪微服务的版本，并解决其路由问题。
• 微服务会消耗比单体应用程序更多的资源。虽然我提到的优点之一就是可以更出色、更有效地利用可扩展性和资源，但是所有组件都需要有自己的实例和容器，这可能就会导致内存和 CPU 使用量增多。

Growing Hacking Strategies and Techniques in Example

Growing hacking is becoming more and more popular in the filed of marketing. It effectiveness is believed to reflect in the company whose size from start-up to small/middle-sized. The right strategy plays as the role of catalyst to drive the high-speed growth of the company in the early stage. The appropriate techniques which are adaptable along with the business development is another key element to ensure that the hacking strategy is applied in the right spot of the whole business. 

John Mcelborough's article in his blog summarise a total 51 examples of growth hacking strategy and techniques. A very deep analysis article. There should be no best growth hacking strategy and techniques for all. But the good ones serve as the fuel of the business engine. 

Andrew Ng's AI Efforts in Manufacturing

Revitalizing manufacturing through AI

Dear Friends, 

I am excited to announce Landing.ai, a new Artificial Intelligence company that will help other enterprises transform for the age of AI. We will initially focus on the manufacturing industry.

AI is already transforming the IT industry. In my work leading Google Brain and Baidu’s AI Group, I’ve been fortunate to play a role in the transformation of two great Internet companies, and see firsthand the benefits modern AI brings to these businesses and to their users. It is now time to build not just an AI-powered IT industry, but an AI-powered society. One in which our physical needs, health care, transportation, food, and lodging are more accessible through AI, and where every person is freed from repetitive mental drudgery. For the whole world to experience the benefits of AI, it must pervade many industries, not just the IT industry.

AI transformation is hard

Many companies are figuring out how to use AI, but this is not easy. The technology is still complex, and few teams understand AI well enough to implement it effectively. Outside the IT industry, almost no companies have enough access to AI talent.

Further, just as using IT to transform a traditional company requires more than building a website, using AI to transform a company requires much more than training a few machine learning models. The strategy of integrating AI — everything from data acquisition, to organizational structure design, to figuring out how to prioritize AI projects — is as complex as the technology, and good AI strategists are even rarer than good AI technologists.

Landing.ai will help enterprises address these challenges. We are developing a wide range of AI transformation programs, from the introduction of new technologies, to reshaping organizational structure, to employee training, and more.

The Manufacturing Industry

The IT industry has primarily shaped our digital environment. Manufacturing touches nearly every part of our society by shaping our physical environment. It is through manufacturing that human creativity goes beyond pixels on a display to become physical objects. By bringing AI to manufacturing, we will deliver a digital transformation to the physical world.

AI technology is well suited to addressing the challenges facing manufacturing, such as variable quality and yield, inflexible production line design, inability to manage capacity, and rising production costs. AI can help address these issues, and improve quality control (see our video demo), shorten design cycles, remove supply-chain bottlenecks, reduce materials and energy waste, and improve production yields.

We are also excited to announce today a strategic partnership with Foxconn. We have been collaborating with Foxconn since July, and are developing AI technologies, talent and systems that build on the core competencies of the two companies. As one of the world’s leading technology service providers and a multinational company running manufacturing in several continents, Foxconn provides Landing.ai a platform to jointly develop and deploy AI solutions and training globally.

Jobs and training/retraining

Bringing AI to manufacturing will also revitalize manufacturing jobs in the US and globally. There has been much discussion about how people will work in an AI-powered future. The next wave of manufacturing jobs will be very different than the previous one. They be higher-level and higher paying, but also require new skills. Thus, they will require large scale training or retraining.

The retraining of workers for the next generation of jobs in AI is a challenge that my team is uniquely equipped to tackle. We are dedicating time and resources to creating retraining solutions for current or displaced workers. We are discussing the deployment of skills training programs with a variety of partners, including local governments. We hope we will have more to say on this soon, and are excited to take on this challenge.

In developing economies, the AI transformation of manufacturing will accelerate the affordability of products ranging from antibiotics to bicycles to computers. It will also help small-scale producers sell products to and benefit from global supply chains. In developed economies, deeply integrating AI into manufacturing will also pave the way to power a new generation of products, devices and experiences.

If you are interested in learning more, please visit our website landing.ai or contact us at contact@landing.ai.

We look forward to sharing more about Landing.ai in the near future!

Andrew Ng

CEO, Landing.ai