I stumbled upon Paul Graham's (fedi) 2002 article "A Plan for Spam", which apparently was the start of the wide usage of Bayesian algorithms for spam filtering.
I'm just a receiver of spam and user of such software but I found this a very interesting read. And the optimistic look into the future, the anticipation of the near-eradication of e-mail spam felt good.
It made me realise how good spam filtering actually has become/how much worse it has been 25 years ago and that my only problem with it is the lack of a good server-side (mail client independent) implementation of what we know to be working; with my mail hoster that is.
For 20 years I've seen it mentioned again and again that a spam filter trained on your good and bad mail will separate future incoming mail well if the data set is large enough. But nobody ever explained to me how that is done. I wish somebody would have just pointed me to this article earlier because it's really interesting and not as complicated and difficult to understand as I assumed.