According to Linux Weekly
News, Eric Raymond is writing a new spam filter called
based on Bayesian
analysis, as suggested by Paul
Graham. Unlike the excellent SpamAssasin, which merely requires
whitelisting a small number of addresses,
training with around 1,000 e-mail messages. But
ultimately offer more hope for defeating spam.
bogofilter recognizes most incoming spam
(allegedly as much as SpamAssassin, but we'll have to wait and see).
More importantly, however,
bogofilter is very good at not
recognizing legitimate e-mail as spam (in other words, it has a very
low false positive rate).
The secret strength of
bogofilter, however, is the training
process. Because bogofilter is trained by the user, each user gets a
personalized spam filter. This means that (1) information of
professional interest to the reader will generally be recognized as
non-spam (however incriminating it might otherwise look), and (2) there
won't be a centralized
list of rules for the spammer to read.
I suspect that the new MacOS X 10.2 mail client may be using a similar technique.