Learning in Public: 3/24/2021

Interesting bits and bytes

Bill Ferrell

Mar 24, 2021

Share Links and Learnings

Not Born Yesterday by Hugo Mercier

I do not recommend this book.

Notes:

People are really hard to convince / sway on issues where they are knowledgable AND OR where they stand to benefit/lose significantly.
Most often, when we look for an explanation we are post hoc explaining / rationalizing. So blaming FaceBook, Trump, people being gullible — really that is just us looking for the “easy” explanation.
1. And this is why this book was so unsatisfying. It gave the complex answer — when, what I wanted was the easy answer.

Quick Hits:

WTF is Crypto Art / NFTs (non-fungible tokens). Here is an argument for Crypto art: https://scottbelsky.medium.com/the-furry-lisa-cryptoart-the-new-economy-of-digital-creativity-6cb2300ea081
What are you buying when you buy crypto art? Answer: Not much. Something that will almost certainly not be around in 100 years, probably not around in 5 years. Read more:
Jonty Wareing @jonty
Out of curiosity I dug into how NFT's actually reference the media you're "buying" and my eyebrows are now orbiting the moon
March 17th 2021
7,394 Retweets27,621 Likes
And a good rundown on why NFTs (non-fungible tokens) are hard to explain: https://medium.com/@nic__carter/why-nfts-are-hard-to-explain-48f0ab0a35bf
Last bit on CryptoArt/NFTs: My previous post is now available to buy as CryptoArt: https://mintable.app/collectibles/item/Wasaucecom-Learning-in-Public-3212021-httpswferrellsubstackcomplearning-in-public-3212021/ZB46qHMEcw42Etv
Image Similarity Search
Imagine you wanted to build image similarity search 10 years ago — e.g. I have a photo of a car, find more photos of cars like this. I would have said it’s possible but months of work. 20 years ago, I would have said — that is a task for a company requiring many many people and significant financial resources. Today, it’s something you can build in an afternoon: https://www.pinecone.io/learn/image-similarity-search/
Companies that make it easy to build, train, test and serve machine learning models are great examples of taking on undifferentiated heavy lifting — and something I think we will see broad adoption for over the next few years. Most companies want the benefit of machine learning, and soon will all have a small machine learning team — but they don’t want those women and men working on infrastructure, they want them developing better models to move the business metrics. Exciting that all the complexity of leveraging ML models will soon be hidden behind an API call.
Comparing Covid vaccines:
Efficacy measures relate to WHEN, WHERE and WHO were included in the vaccine study. This graphic below does a great job showing how the vaccines didn’t measure efficacy on equal grounds — J&J testing when the virus was much more actively spreading.
More details and insight:
Buying Geoff Hinton - one of the "Godfathers of AI". Great story of how Geoff auctioned off his talent and research lab to Google: https://www.wired.com/story/secret-auction-race-ai-supremacy-google-microsoft-baidu/
Scripto.live - Stephen Colbert’s software product. Building software for a specific problem in scripted TV. What a great example of building a business around a real business problem — not building software just to re-build the world in software.