The Five Best Things: Oct 3, 2020

Slow news week, eh?

Hi all! Highlighting a couple of interesting ML topics this week. The honorable mentions list is extra long, because I’ve been seeking out good news as a distraction from the madness of our times.

I’ve also created a “Job Drop” section to pass along neat hiring opportunities in my network. I hope this is helpful, and if you have job openings you’d like to share with this group, please pass along. We are at 400 subscribers and counting!

The Five Best Things

  1. MIT Technology Review: This know-it-all AI learns by reading the entire web nonstop

    • The article covers a company called Diffbot, which trawls the web and vacuums up factoids using Natural Language Processing and Computer Vision, and turns them into an interconnected network of facts. This is known as a knowledge graph - think webs of subject-verb-objects. While knowledge graphs are not a new concept, they’ve been hard to build and keep fresh in an automated fashion; prior to Diffbot, only Google and Microsoft have pulled off crawling the entire web.

    • The article presents Diffbot in contrast to OpenAI’s new GPT-3 175B parameter mega-model (more on GPT-3 in a future edition). I’m very excited by what a freely available, constantly updating knowledge graph will allow us to do. It may even be more useful than GPT-3, which garners the lion’s share of headlines. Snapchat uses Diffbot to extract highlights from news pages; wedding-planner app Zola uses it to help people make wedding lists; NASDAQ uses it for financial research; Adidas and Nike use it to search for counterfeit shoes. An example of putting in lots of hard work for long term payoffs.

  2. Stitch Fix Tech Blog: Multi-Armed Bandits and the Stitch Fix Experimentation Platform

    • The Stitch Fix data science teams presents an emerging alternate technique to A/B testing, called multi-armed bandits. Multi-armed bandits, which borrow concepts from Reinforcement Learning (RL), are similar to A/B tests in that they randomly allocate traffic to different variants of a web page or algorithm in order to determine which one results in better outcomes. MAB differs from A/B tests by diverting traffic away from poorly performing variants to better performing ones while the experiments is running, thereby reducing the opportunity cost of experimentation.

    • A/B testing is a widely-used randomized experimentation methodology to infer if users respond in a statistically significant way to changes in the user experience. However, one cannot infer the results of these tests until after the experiment concludes, which may be too late, and/or result in lost revenue. The multi-armed bandit is a simplified RL agent to get around this problem. The tradeoff is that it can be much harder to analyze statistical significance, or control error rates. Still, this is a good tool to have in a marketer’s toolbox.

      visual of ab testing vs bandit for different types of program


  3. Invest Like the Best Podcast: Matt Ball – The Future of Media: Movies, the Metaverse, and More

    • In this 105-minute podcast, Patrick O’Shaughnessy and Matthew Ball go into all things entertainment, covering Netflix, Disney, Spotify, Epic Games, Tencent, and Cloud Computing’s role undergirding all of these. The commonality across these is a) empire building and b) positive feedback loops, which are germinated, nurtured and protected over extraordinarily long periods of time, providing these companies unimpeachable moats in their respective arenas.

    • I love following Matt Ball’s work, because he helps me gain an insight into creativity and gamer culture that I lack from first hand experience. This further helps me understand the complements to these trends, such as the chips and networking gear provided by Nvidia, and cloud computing. Additionally, Tencent is a company I became massively interested in after conducting research into eSports for a class project in B-school, involving taking Gamestop private. If you have some time to spare, here’s the report I worked on with my wonderful teammates Emily, Colin and Manasa. IMO, Tencent + Nvidia are going to take a cut of the gaming software and hardware markets for years to come.

  4. Wall Street Journal: Inflation Is Already Here—For the Stuff You Actually Want to Buy

  5. Wall Street Journal: Forget the Stock Market. The Rare-Plant Market Has Gone Bonkers.

    • Another article on inflation… in rare plants! With everyone confined to their homes and posting to Instagram, returns on unusual plant varieties are approaching 10x. Variegated monstera adansonii prices have gone from double digits to the $2000-5000 range.

    • How can you not love an article with a quote such as this - “Does your Chanel handbag grow another Chanel handbag in a month?”. Related, the black girls with gardens account has been my favorite Instagram follow of late.

Honorable Mentions

Job Drop

Mobile UX Designer, Flowly - Pasadena, CA/Remote