The Five Best Things: Oct 31, 2020

Sweet and scary things on Halloween!

Happy weirdest Halloween of our lives!

I should have made it more obvious I was going for the Barbra Ann/GAN joke last week :-)

Do check out the Job Drop for a role in my org at Google, and let me know if this fits your experience and interest!

I’ve written about AMD in the past; they announced blockbuster earnings this week, with huge growth in the Datacenter segment, in contrast to their rival Intel. And in continuing empire building in the semiconductor industry, announced their intent to acquire FPGA company Xilinx.

Why I find this interesting: Xilinx’s SmartNIC product line is designed to reduce the reliance on Intel’s CPUs in a data center environment, so the bad news for Intel is likely to continue.

The Five Best Things

  1. This video from GMA

    • Ben Einstein co-founded BoltVC, one of the few firms that still invest in hardware. He was let go in a very murky manner in 2019, when his wife was at the beginning of her cancer fight. He’s now an Entrepreneur-In-Residence at Eclipse Ventures. This story is beautiful and I hope it relieves some of the anxiety we are all feeling at the moment.

  2. Michael Bronstein, Head of Graph ML research at Twitter: What Twitter learned from the Recsys 2020 Challenge

    • Twitter sponsored the Recommendation Systems challenge at RecSys this year, by providing a dataset consisting of tweets and user engagements over a 2 week period. The competition asked participants to build models to predict the probability of a user engaging with a tweet. Nvidia won this year’s challenge by virtue of parallelizing the workload - allowing them to complete training the model in 2 minutes vs hours on alternate hardware. The model uses extreme gradient boosted (XGBoost) trees and adversarial validation (aka GANs) for feature importance and selection.

    • Recommendation engines, the models that suggest what you should buy next on Amazon or watch next on Netflix, are a HUGE business. For context, 35% of Amazon’s revenue - nearly $100B - is credited to its recommendation engine.

    • These models tend to use an algorithm called collaborative filtering. What I liked about the winners of this competition was that deep learning was rarely used - indicating that either they are a bad fit for this insanely lucrative market, or that much more work needs to be done. The Nvidia paper implies that there isn’t enough information to be gleaned to justify the use of the more computationally expensive BERT-based deep learning models. I’m stunned at the speed advantage Nvidia was able to achieve - 2min vs hours on CPUs. That’s real impact and dollars.

  3. Packy McCormick: Knock Knock. Who's There? Opendoor.

    • In this persuasive piece, Packy McCormick analyzes the potential of Opendoor, an online real estate transaction company, which allows people to buy a house with the click of a button. It was recently taken public via Chamath Palihapitiya’s $IPOB SPAC. He argues how the company is similar to Amazon in its early days, by emphasizing vertical integration, big data, value added services, and long term growth over short term gain. By slowly eliminating the structural risks in the last few years, it’s poised for outsize growth.

    • I was highly entertained this post, which combines both emotional persuasion and logical reasoning about what would appear to be a crazy, overinflated and overtly risky idea even in today’s times. But then again, that’s what folks said about Amazon, whose culture, values and flywheel dynamics are similar to Opendoor’s. This stuck with me - “IPOs are for easily-understood companies, and SPACs are for companies whose potential is underappreciated.” I was surprised to learn the company actually has positive contribution margins, which is rare for a young company in the current environment. It will be interesting to watch them as a public company.

  4. Gene Weingarten: A neighbor asked for a tomato. This is where the story gets weird.

    • A short story from Gene Weingarten covers a strange series of interactions with his neighbor. It starts off friendly and wholesome, and culminates with an unexpected twist that redefines neighborly behavior!

    • A short, engaging story with a twist that comes out of nowhere. Enjoy on Halloween night!

  5. Mundane Halloween in Taiwan and Japan

    • In 2014, the Japanese started a tradition of "Jimi-Halloween”. It was started by a group of people who “kind of wanted to participate in the festivities of Halloween, but were too embarrassed to go all out in witch or zombie costumes.” It’s now turned into a huge event and spread to other countries like Taiwan. By the way, Japan reported 770 cases of Covid today, Taiwan reported 1, and the USA reported 99,784.

Honorable Mentions

Job Drop

  • Product Manager, Google Cloud - my org is looking for folks with experience in the HPC/Supercomputing space particularly, and is open to grooming engineers wanting to make the transition to product. Please reach out if you fit this criteria.

Stay safe everyone!

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