[Spambayes] SpamBayes as a gatway solution

[ROTTENBERG,HAL (HP-USA,ex1)]

I saw an article the other day (freshmeat.net review of spam filters) that
mentioned the position of the author that spam filtering at the server level
should only consist of "checking outside of the envelope", not "opening the
envelope".  In other words RBL and obviously fudged mail headers.  And leave
the filtering of the content side to the client.  For CPU reasons surely
that is a valid point.  

But can we do better?  What about installing clients on everyone's desktop,
but the corpuses (corpusi?  :) will feed in to a master server, and it would
do comparitive score analysis across the results and only keep the worst of
the worst.  Even if that only catches 10% of the spam, that would save
network bandwidth.

Hal Rottenberg
Windows Delivery Engineer
Hewlett-Packard
404-774-4041

> -----Original Message-----
> From: Peter Beckman [mailto:beckman at purplecow.com] 
> Sent: Wednesday, August 27, 2003 2:17 PM
> To: David Bear
> Cc: Itamar Rosenn; spambayes at python.org
> Subject: Re: [Spambayes] SpamBayes as a gatway solution
> 
> On Wed, 27 Aug 2003, David Bear wrote:
> 
> > On the other hand, if you stick with the vigilant and conservative
> > system admin, you have something like "The Spammish Inquisition".
> > Great idea..  A filter on the MTA that will 'slow' or sqelch the
> > bandwidth a spammer has.  Now, thats a powerfull idea.
> 
>  The thing I don't like about that is the fact that a legitimate mail
>  server may get DOSed by your mail server if your filter is 
> wrong.  You may
>  even take someone elses mail server out, because yours can 
> handle more
>  open connections; what happens if you are wrong?  The people you are
>  "squelching" can't contact you to tell you that you are 
> "squelching" them
>  incorrectly because you are "squelching" them.
> 
>  I think SpamBayes implemented by each user using a common 
> code-base and
>  forcing them to teach SpamBayes individually would be the 
> best Bayesian
>  solution.
> 
>  SB has given me much better results than SpamAssassin as of yet.
> 
> Beckman
> --------------------------------------------------------------
> -------------
> Peter Beckman                                                 
>  Internet Guy
> beckman at purplecow.com                             
> http://www.purplecow.com/
> --------------------------------------------------------------
> -------------
> 
> _______________________________________________
> Spambayes at python.org
> http://mail.python.org/mailman/listinfo/spambayes
> Check the FAQ before asking: http://spambayes.sf.net/faq.html
>