I contend that this isn’t true.
First of all, if it was true, there would not be so many deliverability problems with email.
But, more importantly, there are solid, definitional reasons why spam filtering can never be particularly good. Is it spam if a bot copies questions from one mailing list to another one with a similar topic, in an attempt to build out a reputation? The copied “spam” emails have identical content to legitimate traffic, but there’s no human at the other end, and the seemingly-good user is eventually going to “cash in” on their reputation by posting actual spam. Similar tactics can attack Bayesian filters, and IP address-based systems are getting less useful over time as CG-NAT becomes more popular.
Personally, I think “spam filtering” as a heuristic that you tack onto an otherwise spam-vulnerable protocol has been about as good as it will ever get for the last decade [1].
Email and traditional web forums share most of the same problems, but web forums have two big advantages:
* Forums can reactively remove spam, as can USENET [2]. Mailing lists have to filter everything before fanning out. ML archives can be redacted, but anyone maintaining their own personal archive has to do any post-delivery filtering themselves.
* Forums have somewhat more operational flexibility. You won’t see Reddit/HN-style voting in mailing lists unless you customize your local tooling to support it. Normally, all a ML can do is decide whether to block it entirely or not.