Hello Interconnected Readers:

Didi’s expansion plan in Mexico seems to follow me around (or perhaps I’m the one doing the following). I’ve been spending the last few days in a city north of Mexico City in the Bajio region, Queretaro, to meet local entrepreneurs. Queretaro is a city few outside of Mexico has heard of. Yet, on my very first meeting at a local posh coworking office, lo’ and behold, Didi is there too! Its offices are so new the desks and chairs were just assembled.

Didi's office in Queretaro. Yes, the reflection in the mirror is me!

[If you’d like to skip ahead to the six news stories and commentaries – three from English language sources, three from Chinese language sources -- here are this issue’s items:

Didi’s foray into the LATAM region started in January 2018 with its acquisition of 99, a Brazilian ride-sharing company. In April of 2018, Didi entered Mexico with its bread-and-butter ridesharing service. It wasn’t until April 2019 that Didi launched its food delivery service, Didi Food, in Monterrey and Guadalajara -- two of the largest cities in Mexico. As I shared in last week’s Interconnected Weekly, Didi is investing and growing aggressively in Mexico, by offering earning incentives for drivers and cheaper services for lower income users to grab market share. What’s most noteworthy here in Queretaro is that besides Didi Food, it’s also promoting its payment gateway product, Didi Pay.

If you ask any one who is anyone in the Mexico tech scene, you’d soon learn that fintech is the hottest area of startup activity and investment. According to a report by Finnovista, there are 441 fintech startups in Mexico. And this landscape below does not include the fintech products of big players, like Rappi and Didi.

I will write a more comprehensive deep dive of what I learned about Mexico’s tech entrepreneurship ecosystem after I finish my trip in early March. But at this point, it’s safe to say that Didi’s expansion is serious, costly, and at least when it comes to my itinerary, omnipresent.


Before you read on, please check out last week's deep dive post: "Where Will the Next 50 Million Developers Come From?"

I helped build ByteDance’s vast censorship machine” (English Source: Protocol)

My Thoughts: This piece by Shen Lu of Protocol is important and well crafted. It is a first person account of a ByteDance engineer who’s involved in building the company’s censorship machine and reads almost like a confession in a diary. In some sense, what’s revealed is not that different from what Facebook, Youtube, and other US social media companies do when it comes to content moderation. The playbook and organizational structure is quite similar, where some engineers and product teams are devoted to building algorithms based on large datasets to automate the detection of unwanted content, supporting contractors who live in lower-cost places to do the manual moderation work. For Facebook and its peers, it's the Philippines and similar regions. For ByteDance, it’s lower-tier cities in China, like Chengdu and Jinan. Of course, the most important difference is while political speech is not censored and even somewhat algorithmically encouraged on Facebook (because it drives so much engagement and attention), it’s a red flag bluntly applied across all platforms in China with few negative consequences to the regulators. When Clubhouse was shut down in China a few weeks ago, many people asked “why?”, but the right question to ask if you want to see the issue from the Chinese regulators’ perspective is: “why not?”. There’s a ton more we can discuss, but for this weekly issue, I’ll just share a few passages from this article that to me paints a nuanced picture of content moderation in China:

“The truth is, political speech comprised a tiny fraction of deleted content. Chinese netizens are fluent in self-censorship and know what not to say.”

“We would openly talk from time to time about how our work aided censorship. But we all felt that there was nothing we could do.”

“During livestreaming shows, every audio clip would be automatically transcribed into text, allowing algorithms to compare the notes with a long and constantly-updated list of sensitive words, dates and names, as well as Natural Language Processing models.”


"But in China, one primary function of these technologies is to censor speech and erase collective memories of major events, however infrequently this function gets used. "

Can Computer Algorithms Learn to Fight Wars Ethically?” (English Source: Washington Post)

My Thoughts: This is a worthwhile, longform piece on the American military’s attitude towards applying AI on the battlefield. The general impression I got from reading this article is that the top brass of the Pentagon are more and more receptive to letting AI take the reins and make decisions. This attitude is partly due to the rising human costs of the long military engagements America has been having in Iraq, Afghanistan, Syria, and other regions. Another dimension is the real or perceived notion that America is losing the race to AI dominance vis-a-vis China and Russia. One passage I found troubling is:

“The only rules for autonomous military weapons themselves were written a decade ago by a mid-level Pentagon official trying to imagine computer capabilities that were just beginning to seem plausible.”

As many who work in the AI industry know, what’s been driving AI’s advancement in the last decade is Deep Learning. It looks like the rule makers have not done any updating to account for this advancement. While Deep Learning has been more effective than just about any other approach, why it does what it does is oftentimes hard to understand and reproduce. That makes ethical application extremely hard to both prove (mathematically) and enforce (legally).

China Blocked Jack Ma’s Ant IPO After Investigation Revealed Likely Beneficiaries” (English Source: Wall Street Journal)

My Thoughts: Ever since Jack Ma’s Bund Finance Summit speech on October 24, which I translated, the story around Ant IPO just keeps getting more intriguing. The initial media narrative of -- Jack Ma spoke out of turn so the Chinese government showed him who’s boss -- was too simplistic, and I wrote a piece deriding it as much. This new report uncovered another angle to the story -- one about preventing future political rivals from gaining wealth and power from Ant’s IPO. One of these rivals is the grandson of China’s former president, Jiang Zemin.

Xiaomi decides to build cars: Lei Jun may lead the initiative personally, seen as a strategic decision” (Chinese Source: Tencent Tech News)

My thoughts: Xiaomi’s founder and CEO, Lei Jun, is famous for coining the phrase: “if you stand where the wind blows, even a pig can fly”, tying entrepreneurial success to knowing and riding waves. It looks like the next wave he is trying to ride is electric vehicles. This article explains some of what that wave tangibly means in China -- guaranteed investment from government-affiliated funds, incentives from local governments to provide cheap land to build manufacturing facilities as long as there’s job creation. However, China’s own EV market is already getting crowded with NIO, Li Auto, XPeng, BYD, and of course Tesla. Baidu also plans to make its own EVs within the next three years. Has Le Jun missed the EV wave already?

With market cap back to $100 billion, what do investors think of Baidu's stock price turnaround?” (Chinese Source: 36Kr)

My thoughts: Speaking of Baidu, its stock price has been soaring in the last few months as the company eclipses the 100 billion USD market cap milestone. I have not been terribly impressed with Baidu as a company, given its questionable practices in its search business and inability to attract top technical talent compared to other Chinese Big Tech’s. However, as the author of this article analyzed, the stock market is validating its narrative as a tech company with huge growth potential. The three components of this growth story are: Baidu’s own AI chips, Mobility-as-a-Service (MaaS), and its own EVs, much of the software side will be powered by its own self-driving open source project, Apollo. While it’s too early to tell if any of these initiatives will come to fruition, from at least an investment angle, looks like I missed the Baidu wave.

Ant Design’s repo got deleted!” (Chinese Source: AI Tech Camp)

My thoughts: While many people, including myself, have been generally critical of Chinese tech companies doing more “using” than “making”, when it comes to open source technology, that trend is starting to shift. One recent example that made some headlines is Ant Group’s open source project, Ant Design, which is a UI design framework based on React. Its repo suddenly disappeared on GitHub last week, causing a minor panic and users asking about it on StackOverflow. It’s now restored. The project is popular and widely used, with more than 1,300 contributors and over 185,000 GitHub users whose projects depend on it.

If you like what you've read, please SUBSCRIBE to the Interconnected email list. To read all previous posts, please check out the Archive section. New content will be delivered to your inbox (twice per week). Follow and interact with me on: Twitter, LinkedIn, Clubhouse (@kevinsxu).

Hello 《互联》读者们:




滴滴从2018年1月开始进军拉美地区,首先以收购巴西的一家共享乘车公司99为第一炮。2018年4月,滴滴进入墨西哥,推出了自己核心的共享乘车服务。2019年4月,Didi才在蒙特雷和瓜达拉哈拉两座墨西哥大城市推出外卖送餐服务,叫Didi Food。正如我在上期的《每周互联》中分享的,滴滴正在墨西哥积极投资和发展,通过给司机加薪和为低收入用户提供更便宜的服务来抢占市场份额。在Queretaro这里最值得一提的是,除了Didi Food,它还在推广其支付产品,Didi Pay。





我帮助建造了字节跳动庞大的‘删帖’机器”(英文来源: Protocol)

我的想法: Protocol的沈璐的这篇文章含义很深,写的也很精巧,值得一读。它是一位字节跳动工程师的第一人称描述,揭露给公司打造内容审查机制的工作,读起来像一篇个人日记里的检讨书。从某种意义上说,所揭示的内容与Facebook、Youtube等美国社交媒体公司在内容监管方面的做法并无太大区别。其玩法和组织结构颇为相似,某些工程师和产品团队致力于构建基于大数据的算法,以自动察觉所谓不良内容,协助住在成本较低地方的包工做人工审查。对于Facebook及其同类公司来说,是菲律宾和类似地区。对于字节来说,是中国的非一线城市,比如成都和济南。当然,最重要的区别是,虽然政治言论在Facebook上没有被审查,甚至在某种程度上被算法“鼓励”(因为流量很大,关注度很高),但在中国的所有平台上,任何政治言论都是个问题,“删帖”对监管部门也没什么负面影响。几周前,当Clubhouse在中国被封时,很多人问 "为什么?",但如果你想从中国监管机构的角度看这个问题,更该问的是,"为什么不?"。这个问题我们从很多方面继续讨论,但做为本期周刊的六条新闻之一,我只想先分享这篇文章中的几段话,在我看来,很好的描述的内容监管背后细致微妙的情况:






算法能学会怎样有道德地打仗吗?”(英文来源: 华盛顿邮报)

我的想法: 这篇长篇文章讲述了美军在战场上对使用人工智能的态度。读完这篇文章,给我的总体印象是,五角大楼的高层越来越接受允许AI在战场上做主导,做决策的这一未来。这种态度部分原因是美国在伊拉克、阿富汗、叙利亚等地区的长期驻军带来的巨大的人力成本。另一层面则是美国害怕和中国或俄罗斯比,已经开始失去AI主导能力的这一说法,无论是真正的失去,还是感觉上的失去。文章中有一段话令我担忧:


很多在AI业界工作的人都知道,在过去十年里,推动人工智能飞速发展的是深度学习(Deep Learning)。看起来五角大楼的监管官员并没有根据最近的发展做出任何更新。虽然深度学习比其他任何方法效果都更好,但它为什么会出什么结果,很多时候很难理解和重现。这使得道德应用在证明上(靠数学)和执行上(靠法律)上都非常困难。

政府阻止马云的蚂蚁上市,是因为调查后发现背后其他可能的受益者”(英文来源: 华尔街日报)

我的想法:自从10月24日马云在外滩金融峰会上的演讲后,围绕蚂蚁上市的故事越来越耐人寻味。最初媒体的解释和说法 -- 马云嘴大,所以中国政府禁止蚂蚁上市,给他点颜色看 -- 过于简单化,我也为了这个问题写了篇文章。这篇新报道挖掘了故事的另一个角度,关于防止未来的政治对手从蚂蚁上市中获得财富和权力。这些竞争对手之一则是中国前主席江泽民的孙子。

小米确定造车:或由雷军亲自带队 视为战略级决策” (中文来源: 腾讯科技)

我的想法:  小米创始人兼CEO雷军一句名言是:"站在风口上,猪都能飞起来",将创业成功与知道浪潮、知道风口在哪里绑在了一起。看似他要站的下一个风口是电动汽车。这篇文章解释了这股浪潮在中国发展的实际意义 -- 政府下属的基金会大量投资,地方政府也会积极提供廉价土地来建造生产设施,只要能创造就业机会。然而,中国自己的电动汽车市场已经越来越拥挤,有蔚来、理想、小鹏、比亚迪,当然还有特斯拉。百度也计划在未来三年内做出自己的电动车。雷军想站的这个电车风口,是不是已经有点晚了?

市值重回千亿美金,投资人如何看待百度股价翻身?”  (中文来源: 36氪)

我的想法:  说到百度,其股价最近飙升,市值已经突破了1000亿美元的里程碑。鉴于百度在搜索业务上的一些不良行为,加上与其他中国科技大厂比,一贯吸引不到顶尖技术人才,我一直不是很看好百度这家公司。然而,正如这篇文章的作者所分析的那样,股市正在验证百度作为一家具有巨大增长潜力的科技公司的这一说法。这个增长故事的三大要素是:百度自造的AI芯片、Mobility-as-a-Service (MaaS),和自造的电动汽车,其中软件层面大部分将由内部研发并已开源的无人驾驶项目Apollo来驱动。虽然现在判断这些产品能否成功还为时过早,但至少从投资股市的角度来说,看似我也已经错过百度这个风口了。

Ant Design 遭删库!”  (中文来源: AI科技大本营)

我的想法: 虽然业界内的很多人(包括我)都批评过中国科技公司在开源技术方面 "用" 多于 "造",但这种趋势正在开始转变。最近一个上了些头条的例子是蚂蚁集团开源的项目,Ant Design,是个基于React的UI设计框架。它在GitHub上的repo上周突然消失了,引起了一阵小恐慌,促使用户在StackOverflow开始询问,现在已经恢复。该项目很受欢迎,应用广泛,有超过1300个贡献者,超过18.5万GitHub用户的项目依赖它。