The filtering and permissions are done at the work station level, not at
the server level, but I agree that there's still work to be done.

Still, blackhole lists just aren't hacking it (no pun intended). They do
a pretty credible job of blocking spam, but an abominable job by
blocking legitimate email.

Take a service like Microsoft's bCentral which is on several blackhole
lists. Thousands of people use this service legitimately for newsletters
and other opt-in lists, but even Microsoft will admit that the service
is occasionally used for spam, mainly by nonprofessional "spammers"
who've bought some mailing lists. Well over 99% of accounts using this
service are legitimate, but it remains blacklisted.

When 100% of emails from a service are blocked and 99% of them are
legitimate, that's a pretty inefficient system.



-----Original Message-----
From: Armando WarpKat Ortiz [mailto:warpkat@...]
Sent: Friday, June 06, 2003 8:04 PM
To: iiaavu@...
Cc: warpkat@...; coach@...;
i_did_not_get_my_email@yahoogroups.com
Subject: RE: [i_did_not_get_my_email] Critics set up anti-SPEWS Web site

> The Bayesian filtering theory is good for short term, but as email
upon
> email is received and the Bayesian, as I understand it, does its work,
word
> statistics go up and will be a very big problem as legitimate email
gets
> blocked out because words matched what IT "thinks" is "spam" when the
email
> itself could be a very long newsletter containing the very same words
at the
> top of the Bayesian "spam identifier list."
>
> =============================
>
> When an email from a legitimate sender does inadvertently get blocked,
then
> you add that address to the "permitted" list. Unless you routinely get
> emails from thousands of people, that's not much work.

Then this approach would only work for individuals who run their own
mail
servers and not any decently sized company. Administering mail alone is
a
tough job as it is without having to sift through everybody's email to
figure
out what is and what isn't spam. That's unsettling in terms of privacy
to the
employee despite what any company has pertaining to such policy.

As per our policies, I won't look at our employee's emails without their
knowledge and only when there are problems with communication between
our mail
server and another mail server where an SBL has absolutely no
intervention at
all.

You can only automate so much where the automation will just simply
cease to
do anything it was intended for.

Bayesian methods need to grow up more and more testing needs to be done
before
it can be widely implemented.

Personally, I see challenge response doing a much better job than
Bayesian
implementations.