about

I made this site because I need a better place to post long form writing that is quick to update. My rough workflow: I post on Twitter while I’m working on my Colab notebooks & GPUs, and then move things here as I go. I also created this site so that I can have a less important place to work out the language and registers for moving this field forward.

Post-training, evals, interp, and stated reasoning vs. real computation — some of the mechanics I’ve been looking at, often leading me down the pretraining rabbithole. And that means down the literature rabbithole, which is where I came from.

Mostly for the love of the game, partly bc it needs to happen.

Elsewhere:

• Twitter: @yok0zuna — DMs open
• GitHub: mytensorsarelonger, keppy
• Email: me@yokozunasan.com