Bayes filter at ISPs

[Adam ENDRODI <borso@vekoll.saturnus.vein.hu>]

I'm considering replacing the current SpamAssasin to a true
Bayasian filter (bogofilter, actually) on a mail server, because
in personal daily usage, it has proven to be a better (faster
and more accurate) solution for me.

I suppose many of you use Bayesian spamfilters at the ISP level.
I'd like to ask how do you teach it to separate ham and spam
correctly?  In particular, how do I select a representative set
of ham and spam?  Is it a good idea to deploy bogofilter for an
entire organization at all?

thanks,
adam

-- 
Am I a cleric?     | 1024D/37B8D989
Or maybe a sinner? | 954B 998A E5F5 BA2A 3622
Unbeliever?        | 82DD 54C2 843D 37B8 D989
Renegade?          | http://sks.dnsalias.net