On Pinduoduo, economics of AI startups, Indian infrastructure, and cutting-edge physics infrastructure

Interesting Links: August 16, 2020

Underdevelopment is never improvised: it is the work of many centuries.
- Nelson Rodrigues, Brazilian playwright

• Fantastic deep dive into Pinduoduo, now China’s second largest e-commerce company, the “Costco meets Disneyland” company.

• Adventures in improving AI economics by Martin Casado and Matt Bornstein at a16z. Great read on problems that AI systems have when the majority of your data lives in the long tail of a distribution.
  • “Supervised learning models tend to perform well on common inputs (i.e. the head of the distribution) but struggle where examples are sparse (the tail). Since the tail often makes up the majority of all inputs, ML developers end up in a loop – seemingly infinite, at times – collecting new data and retraining to account for edge cases. And ignoring the tail can be equally painful, resulting in missed customer opportunities, poor economics, and/or frustrated users.”
  • “Anecdotally, development costs – and failure rates – for AI applications can be 3-5x higher than in typical software products.”
• How does UPI work? (part2 on the transaction lifecycle and part 3 on the settlement process)
  • UPI is India’s instant real-time payment system. It is essentially what the US Fed is trying to do now with FedNow.
  • Many people who hype up UPI or are detractors miss a key point: This is not about convenience per se (though it would appear that way from the perspective of, say, a Google Pay), but about access. About trying to desperately maintain some semblance of ‘standard’ design and improving access.
  • The absurd over-reliance on a cell phone number is clearly a big issue in India in general (not limited to UPI).
• A post about starting a manufacturing unit in India which was subsequently taken down by the author based on a mistake in the initial process by an employee.
  • Do note that this was an application to start a factory on agricultural land. India has a particularly troubled history when it comes to agricultural land & using it for non-agricultural purposes. It’s not clear if much of this process would have been required for non-agricultural land. But it is also the case that most land in India is / can be deemed to be agricultural land.
  • But it is a good read nonetheless. And the Hacker News comments are very interesting to read and showcases our bias when evaluating government systems. But as is often the case, there has to also be some truth in it for the stereotype to exist in the first place. Some paraphrased exmaples of the comments:
    • ‘Government employees in India take those jobs only for alternative revenue streams aka bribes.’
    • ‘It is the same in the Philippines run by corrupt trapo (“filthy rags”) politicians.’
    • ‘Well, these employees had to pay bribes to get that position in the first place. Its turtles all the way down.’
    • ‘Much of the world operates like this, to soak you for as much money as possible before you give up and pay, even if they provide nothing actually useful or meaningful.’
    • ‘As a famous Brazilian writer said: “Underdevelopment is not improvised; it takes centuries to perfect.”’
    • ‘This is why entrepreneurs with low political influence participate in low capital businesses (ie import export). They move into more capital intensive businesses (ie oil refineries) only when they have bought political influence.’
• And in more inspiring, fun news, assembly of ITER has begun. ITER, being built in France, is slated to be the world’s largest fusion reactor. It’s a global collaboration, with seven member entities overall (the EU, China, India, Japan, Russia, South Korea, and the United States). Brief video by DW.