Some good stuff on YouTube

Fun stuff today!

Andrew Kelley Practical Data Oriented Design (DoD) - YouTube. In a deliberate callback to CppCon 2014: Mike Acton “Data-Oriented Design and C++” - YouTube, the creator of Zig talks about applying DoD ideas to the Zig compiler. In brief: “struct of array” instead of “array of struct” can get you a lot of mileage, or in general, looking for anytime there are a lot of the same type of object in memory, and making each object smaller. CPUs are faster at many things that reading from any level of memory, for example math. Other good stuff.

A Systems-Minded Approach to Creating a Music Player Application by Andrew Kelley - YouTube is fun but less focused and useful; one big point about the youtube-dl project, that the project maintainers’ real product is their labor (the rapid updates to combat YouTube’s obfuscation), which they’re choosing to package in a complicated Python toolchain, which is absolutely their choice, but perhaps the project would get more reach if they were to do this in a more isolated and re-usable library, since again their product is not really software — the code is not the thing they’re offering, it’s the ongoing labor to keep the code useful.

Implicit Surfaces & Independent Research - YouTube by Matt Keeter was pretty inspiring. The talk is about implicit surfaces, but he offers some great advice about doing independent research as the final third of the talk, summarized here:

• You can just do things, like read papers, email authors, write software, publish blog posts, submit to journals.
• There is a virtuous cycle between doing things and talking about them.
  • (Do) Learn more about a subject; get better at the craft; recognize promising ideas
  • (Talk) Write blog posts, submit papers, publish demos, talk about your work on the socials
  • (Talk) Meet interesting people; learn about their ideas; synthesize from conversations
  • (Do) Read blog posts; read papers; write software; do experiments
• Blog-driven development is pretty good
  • Stops open-ended work by giving you a concrete target to focus on
  • Narrowly focus on the write up
  • Reminds you what’s cool about the project
  • Sharing work has social benefits
  • You should have a website
• The pet problems approach (Feynman?)
  • Carry around a little set of pet problems, and every time you encounter a new idea, try to apply it to your pet problems
  • Where do you find pet problems?
  • Adopt them (from your advisor, others you respect)
  • Read a lot (papers, blogs) — find interesting people through news aggregators, and subscribe to them through RSS
  • Look for personal-scale problems
  • Revisit old ideas on modern hardware
• You don’t need to be doing work on the cutting edge of LLMs or whatever; you can make progress wherever you find interest.

This feels closely related to HOWTO-GTD.