Another Interesting View on Bayesian Spam Filtering Effects

Tim Peters, a Really Smart Guy who is responsible for a lot of the most clever Python code and currently (among other things) a developer on the SpamBayes project, posted an interesting comments in comp.lang.python. It shows another possible 'second order' effect of spam that could occur if the odd few people who seem to respond to the spam, and thus may want it, start using Bayesian-type filters to make sure they get that type of spam, and only that type of spam.

In other news, I'm just about to start coding the attack I proposed on November 19th, 2002 tonight. I've designed it, got the classifier in place, and obtained my sample emails. (The 'good' emails are not as diverse as I'd like, but before I go to too much effort to collect more, I'd like to see if it may be enough already... it IS 20,000+ good messages...) So, we'll soon see if I'm full of it or not in a concrete way...