This article was written with the assistance of AI. If that bothers you, stop reading here. The numbers are real: every score comes from the ten training runs described below, and the full run log is in the linked knife.day write-up.

"I didn't write any of this, but you should still trust that the remaining work, ideas and observations are all mine."

To include a disclaimer like this is to fail to recognize that "real numbers" are way less meaningful when there's clear evidence that the prompter of the LLM is not really qualified to validate them.