38
An Analysis of DeepMind's 'Language Modeling Is Compression' Paper
(codeconfessions.substack.com)
This is a most excellent place for technology news and articles.
It looks like they did it both ways (“raw rate” vs “adjusted rate”):