That's fascinating that the total size is so tiny — only 2.4 TB‽

I assume this must be only the text portion, and heavily compressed?

Text really doesn't take up that much space, and in addition it compresses pretty well.

The entire English language Wikipedia is only around 60GB in a format that can be readily searched and randomly accessed (ZIM), for example: https://kiwix.org/

Does Kiwix actually work? I see people hyping it here but I could never get it to actually, y'know, download the file and display the wikipedia on my phone.

It works perfectly for me, both on iOS and macOS.

Just downloaded. Doesn't seem to want to download Wikipedia on my phone, it says "detecting if filesystem supports 4gb files"

Kiwix worked for me. IIRC there may be difficulties opening an archive that was downloaded outside of the mobile app, but archives downloaded in-app were fine.

For the mobile app I used one of the smaller Wikipedia subsets, since I didn't want to take up too much space on my phone. The full offline Wikipedia download is saved to my laptop.

The question is: if you then added all of Usenet before, say, 1992, would the effective intelligence of the trained LLM go up or down?

I can't speak to intelligence, but the result would be ignorant in a meaningful way.