Paul Graham, the creator of Yahoo! Store and other Lisp software, is working on a Web-based e-mail service that detects spam by analyzing word choice in e-mail. The technique requires separate collections of spam and non-spam and works best when personally tuned to your own actual e-mail rather than a generic database of e-mail. "Using a slightly tweaked (as described below) Bayesian filter, we now miss less than 5 per 1000 spams, with 0 false positives," Graham writes. I began filing deleted normal e-mail and spam separately today. When I've amassed at least 4,000 e-mails, I'm going to implement Graham's idea in PocoMail, a scriptable Windows e-mail client I've been using for a while.