> Ultimately, by putting our email in a central place, capturing the text of every email message, and analyzing it, we got a world where you could use email and largely be free of spam. But it did come at a great privacy cost; the same access that allows ML to categorize your email allows anyone to read your email; server administrators, hackers, law enforcement.
I'd just like to offer a counterpoint here. I've been running an email server for myself and some friends for more than five years now and I've been surprised with the low amount of spam we were getting.
Two of the users eventually got caught in a more obnoxious torrent of spam so I configured rspamd. It uses a variety of techniques, including a Bayesian filter score, blackhole lists and greylisting. It resolved all spam issues with zero false positives. Either our spam is an extreme outlier (it doesn't look like it) or filtering spam doesn't require as much ML we're led to believe.
I'd just like to offer a counterpoint here. I've been running an email server for myself and some friends for more than five years now and I've been surprised with the low amount of spam we were getting.
Two of the users eventually got caught in a more obnoxious torrent of spam so I configured rspamd. It uses a variety of techniques, including a Bayesian filter score, blackhole lists and greylisting. It resolved all spam issues with zero false positives. Either our spam is an extreme outlier (it doesn't look like it) or filtering spam doesn't require as much ML we're led to believe.