Gmail has advanced AI-based filtering. Now that LLMs are becoming democratized, I'm ready for a self-hosted AI spam filter.
It seems much more popular to publish research papers on the viability of LLM-based spam filtering than it is to build LLM-based spam filtering software. Here are dozens of papers: https://www.arxiv-sanity-lite.com/?rank=pid&pid=2206.02443 Nothing on GitHub, yet.
Dan G.: to butt in, from experience with #djb from mailing lists for #qmail, #djbdns, etc...
When djb states facts, he is invariably correct. When he presents a logical argument, he is rigorous, but it is frequently non-trivial to follow. Very dense logic, and perhaps omission of the more "obvious" steps, mean you can't grasp it on first read.
When taking apart someone else's argument, it's even denser and crammed with references. But worth it. Spend the time.
Blog post: You should delete the 'User-Agent' header from outgoing email https://utcc.utoronto.ca/~cks/space/blog/spam/RemoveUserAgentHeader
tl;dr: on today's Internet, why would you want to tell other people what mail client you use, and by implication what platform? Strip this out in your mail submission server, and if you can set your mail clients to do it regardless of your mail submission server.
(I don't know how you do this in common mail clients, like Thunderbird.)
It looks like it would be a 1-line patch to new-inject and ofmipd in DJB's mess822. Just drop the header in the very same way that they already drop "Bcc:".
Erwin Hoffmann could probably be persuaded to add it to s/qmail.