Linking Out Loud #24

Hey everyone,

For the foreseeable future, I’ll be sending these every month instead of every week. No particular occasion for scaling back; I’ve just been feeling overloaded, and want to free up some of my time to goof off.

January was a busy month. In addition to the four newsletter issues I sent, I published three blog posts:

I also bought a digital piano and started learning a few songs. Here I am playing a snippet of “Plus tôt”, by Alexandra Streliski:

I finished two books, both excellent:


Being Your Selves: Identity R&D on alt Twitter

20 minutes | Aaron Z. Lewis | 2020

If you’re looking for a way to push the boundaries of your identity, consider making a pseudonymous or “alt” account on Twitter (or whatever platform suits you). If you’ve ever worn a mask, you might have felt that it freed you to behave in ways that would otherwise seem unnatural. In the same way, an alt can be used as a tool for thinking new thoughts. I don’t have an alt myself, but it seems like it would be a fun experiment.

>>> Read it

Noticing You’re Confused

5 minutes | Arram Sabeti | 2020

An eery story about a nearly-averted hiring disaster. The author credits his good decision-making to a lesson from the fan-fiction novel Harry Potter and the Methods of Rationality, a reimagining of Harry’s first year at Hogwarts if Petunia had married an Oxford biochemistry professor, and Harry had been raised as a scientific prodigy. (I’ve read the book twice. Funny premise, but I’d count this book among the five that have influenced my thinking the most.)

>>> Read it

Deep Learning for Symbolic Mathematics

1 hour | Guillaume Lample and François Charton | 2019

A more technical read than usual, but super interesting. It turns out you can train a neural net to compute anti-derivatives and solve differential equations, using the same kind of model that is used to translate English to French. What’s fascinating about this is that unlike natural language, which is often fuzzy, symbolic mathematics is absolutely airtight; there’s no mostly correct answer. As a rough analogy, imagine typing a hard calculus problem into your phone, and seeing predictive text spit out the answer. Importantly, you’re feeding this problem to a system that doesn’t know the first thing about math. It’s just pattern-matching one set of characters to another. Crazy!

>>> Read it


Have a wonderful month!


Did a friend forward this to you? Join 115 others and get it straight from the source: