Micah Johnston

@micahrj.bsky.social

Graphics programmer, Rust enthusiast, and multicellular organism http://micahrj.github.io · http://mastodon.gamedev.place/@micahrj

tip for would-be open source contributors: no matter how good your code change is, if the PR description is a huge brick of subheadings and bulleted lists that clearly came from an LLM, I am going to be fighting the urge to close it immediately and never look at it again

I can feel my brain starting to take damage from the number of times I have started to read something (code or prose) under the assumption it was written by a human and only realized partway through that it was slop

any model-based reinforcement learning algorithm with tunable hyperparameters is secretly just the inner loop of a model-free meta-reinforcement learning algorithm (using any black-box optimization algorithm, e.g. evolution strategies), and the bitter lesson says that you should treat it as such

I think LP5 and EP7 will continue to be my favorite Autechre era, but I'm enjoying all their post-Exai material a lot more than I used to. for a long time it all felt pretty opaque to me