Given the stochastic nature of LLMs and the pseudo-darwinian nature of their training process, I sometimes wonder if geneticists wouldn’t be more suited to interpreting LLM output than programmers.

It is, but maybe they mean they want no limit whatsoever on post length.

which, well, if your instance starts sending out megabyte-sized text posts I don’t expect it to stay federated with many others for very long.

I just searched on GitHub for "Sha1-Hulud: The Second Coming.": 692 repositories. On the first page of results I was able to find a repo clearly made by the malware, and in that repo I was able to find someone’s github token with a few applications of “decode from base64”.

This is pretty bad. I don’t know what exactly comes next, an awareness campaign to get people to clean their infected machines and packages?