[Spambayes] Randomized Spam Beating SpamBayes

[skip at pobox.com]

    Quinn> Is anyone else getting this?  SpamBayes catches all the
    Quinn> "identifiable" spam right off, only a little bit of training
    Quinn> required, but there is a variety that makes it through almost
    Quinn> every time.  It consists of either an entirely random set of
    Quinn> content (no idea what the point to those are) or a small chunk of
    Quinn> content followed by lots and lots of drek.  The subjects are
    Quinn> sometimes coherent, sometimes gobbledygook.

    Quinn> Is there any way to train for random gibberish, or have they
    Quinn> actually beaten SpamBayes?

Well, the random gibberish is probably a prefix to a GIF image containing
the real sales pitch.  Try 1.1a3 and see if that helps.  Note though that
you will probably have to do a little tuning to get it working with Outlook.
I have no Windows machine, so the OCR stuff in the latest version has only
really been tested in a Unix/Mac environment.

Any Windows users want to volunteer to do a thorough test of 1.1a3 under
Outlook?

Skip