Why are we still using a dated GPT-4 benchmark score?

Author here. I've been working on this blog for a little while and am finally publishing the first couple of articles.

It's mainly a way for me to keep track of the many notes and bits of data I come across when reading ML research. I have a lot of content planned, revolving around AI Coding systems, Alignment, and Security.

So today is the first time I'm publishing it anywhere, I can only assume straight into a cold disinterested void ... :)