This is an incredible writeup for really showing the dead ends (the motivation, the effort, and how it turned out), the practicalities (time lost due to random bugs or random things tanking performance), and noting what they don’t know. It’s common not to do any of that, and besides those details being interesting in themselves, the overall picture is encouraging to all us actual humans who hit dead ends, have bugs in our code, and don’t know everything. I love it.
(As a random specific thing, it was kind of neat they found that removing .clone()s in their Rust code had less effect than expected because the compiler was already optimizing copies away for them. In particular, might make me feel a little freer to try cloning my way around a borrow-checker roadblock if I play with Rust and hit one.)
I really enjoyed it for similar reasons. I’m not especially familiar with ML techniques, but at every step in this blog post I was able to go “huh, yeah that sounds like an interesting avenue, I wonder how it will turn out”.
It’s not easy to write like that, so huge kudos to the authors.
Using StickManStickMan images was an especially nice touch.
I had never heard of Simple(x) before and I think I might be able to use it on a problem that I’m working on, so I’m glad I read this.