That’s fascinating.

Why are Big Tech trainer bots crawling so much? Seems horribly inefficient and redundant… don’t they already have archives of most of the net, and couldn’t they just pool resources?

Also interesting that so many folks are still using llama3 8B (unquantized???) and vanilla SDXL.

In other words, it seems much of the worlds traffic from cloudflare’s perspective is just lazy redundancy and seemingly clueless inefficiency…