From: Marc Perkel [mailto:[EMAIL PROTECTED]
Sent: Thu 3/3/2005 10:04
To: [email protected]
Subject: AccuTechnology software patent
I've read your web page and I'm a little confused about what
it is you
hope to patent.
The concept of sebder verification is not
new. In fact it's already in
Exim and Postfix now and I'm using it and it
works great. Unlink what
you are proposing - sending "confirmation" messages
to see if they
bounce - Exim starts a bounce message sequence and sees if it
takes it.
And you can distinguish between misconfigured servers and servers
that
really say that there is no user by that name. I have a number of
Exim
tricks that allow me to get rid of 90% of spam without SA ever
seeing
it. I use SA for the hard ones.
In my setup I bounce incoming
mail that is clearly not from a real user
at connect time. I don't bother to
accept the message and learn it
because these messages are from really poor
spammers and I don't want to
waste the cpu resources to process it. By
rejecting these email it makes
my spam corpus better in that my spam comes
from smarter spammers and is
more of a challenge to detect. Processing
message from dumb spammers
gives you a false sense of success and I think
artificially inflates the
success rates.
As to your superior filter -
are you talking about some new bayesian
like filter you've developed? I am
currebtly running 2 bayesian filters
on my system. The second filter -
spamprobe - is fed custom tokens that
are generated by a program I wrote that
passes only the (enhanced)
headers and hot parts of the body. The output of
spamprobe is turned
into 11 levels of results and fed back into SA for
scoring. Are you
doing something like this? If so - I'm already doing
it.
I'm against software patents in the first place - but I question
from
what you have on your web site if you have anything that really
is
patentable. If so - it doesn't seem to describe it. So - I'm left
confused.
