The Five Best Things: Jan 24, 2020

Memes as far as the eye can see

ICYMI, a picture of Bernie Sanders looking stodgier than usual at the Biden-Harris inaugural has been photoshopped ad infinitum! I’ve gathered my favorite ones after the top 5 reads of the week.

The Five Best Things

  1. Chip Huyen: Machine learning is going real-time

    • Chip Huyen covers the current state of real time machine learning. The definition of “real time” is fuzzy, and Chip helps clarify two distinct kinds of real time ML:

      1. Online predictions is a model making predictions in real time (milliseconds). Examples here include Google search results, high frequency trading, autonomous vehicles. Making the most relevant predictions in the fastest time possible has real life, real $ implications. To speed up real time predictions, ML engineers should look into making the model faster, smaller or run on specialized/latest hardware.

      2. Online learning is a model incorporating new data and updating itself in near real-time (minutes). This is a much harder and more expensive problem; practically this is achieved by updating a model’s weights once a microbatch worth of new samples are collected, evaluating the performance in a sandbox environment, then deploying it wide. TikTok, Weibo, Twitter recommendation algorithms are good examples of this. From Chip’s research, it appears that Chinese firms are at the forefront of online learning, while it is still nascent in American firms.

  2. Byrne Hobart: The Road Not Taken: Stripe, Ant, PayTM & Defi

    • Byrne Hobart’s free weekly post provides the best summary of payments infrastructure and modernization across three large economies of the world - the U.S, China and India. Based on narratives alone, one might expect the U.S payments stack to be the most up to date and modern, China’s to be centrally controlled, and India’s to be a hodgepodge. The reality is much different, and strongly rooted in each country’s history and incentives. Byrne uses Stripe, Ant financial and PayTM as the canonical examples to explain this evolution. Stripe set itself up as the plumbing (via APIs) of the Western financial system, which is byzantine and bizarre. Ant financial, an offshoot from Alibaba, leveraged its ecommerce roots and data mining to become a private financial system outside the CCP’s oversight. PayTM was threatening to achieve Ant-like status (including receiving a $550M investment from Ant itself), until the Indian technocrat bureaucracy came up with the Unified Payments Infrastructure (UPI), an interoperable, real time inter-bank transfer network with open API’s on top of which 3rd party fintech providers build. Thus, arguably the most thriving and modern fintech stack came out of a public-private partnership in India. China is increasingly looking at DeFi (decentralized finance, aka built on top of blockchain) and has launched a pilot program called DCEP.

  3. NYTimes: Underselling the vaccine

    • An interesting discussion of how public health messaging may be failing us yet again in the U.S. We seem to be too focused on cautioning against the longshot side-effects instead of the fact that these vaccines are among the most effective vaccines ever created and are going to get us back to “normal” in record time.

      Here’s my best attempt at summarizing what we know:

      • The Moderna and Pfizer vaccines — the only two approved in the U.S. — are among the best vaccines ever created, with effectiveness rates of about 95 percent after two doses. That’s on par with the vaccines for chickenpox and measles. And a vaccine doesn’t even need to be so effective to reduce cases sharply and crush a pandemic.

      • If anything, the 95 percent number understates the effectiveness, because it counts anyone who came down with a mild case of Covid-19 as a failure. But turning Covid into a typical flu — as the vaccines evidently did for most of the remaining 5 percent — is actually a success. Of the 32,000 people who received the Moderna or Pfizer vaccine in a research trial, do you want to guess how many contracted a severe Covid case? One.

      • Although no rigorous study has yet analyzed whether vaccinated people can spread the virus, it would be surprising if they did. “If there is an example of a vaccine in widespread clinical use that has this selective effect — prevents disease but not infection — I can’t think of one!” Dr. Paul Sax of Harvard has written in The New England Journal of Medicine. (And, no, exclamation points are not common in medical journals.) On Twitter, Dr. Monica Gandhi of the University of California, San Francisco, argued: “Please be assured that YOU ARE SAFE after vaccine from what matters — disease and spreading.”

      • The risks for vaccinated people are still not zero, because almost nothing in the real world is zero risk. A tiny percentage of people may have allergic reactions. And I’ll be eager to see what the studies on post-vaccination spread eventually show. But the evidence so far suggests that the vaccines are akin to a cure.

  4. WSJ: Are Videogames the Future of Remote Work?

  5. Dessa: Who’s Yellen Now

The Best Memes

The internet outdid itself this week with the Bernie at the inaugural photoshops; my favorites -

This whole thread

This one takes the cake though

Honorable Mentions

Disclaimer: The views and opinions expressed in this post are my own and do not represent my employer.