I just threw in SIMD processing on a whim to an existing library.
It wasn't a regular structure that needed one operation or two again and again, it was this irregular object that needed each field each operation to be carefully picked. Assessing what instructions were going to help me do which fields how would have been hard hard hard. Hours of analysis, would have gotten stuff wrong. Would have been so frustrating.
I could just ask for, and I got it.
This is a very specific kind of example. But smart people who know what they want but don't have endless time to go chase every idea themselves will, I think, find great joy getting the AI to go out and land some amazing improvements.
And there's so many projects like zerobrew that just would never be attempted. We can make whole systems, see their performance, and if it's not good, change the stack and generate a new version of the app, a new attempt, with far less effort. We will try many more things.
Kinda all very very obvious to the actual excited doer folks.
I'm curious to how or if this work could be adopted by homebrew, most likely they wrote in ruby(?) since that's what they're productive in.
One approach I find interesting is if we see repos that are actually "specs" for AI to mirror in a more performant language.
I'm sure the core optimisation techniques of this could be done in ruby as well.
reply