we better figure this out…
this is a mess, work in progress
https://openai.com/blog/better-language-models/