Will new filters save us from spam?. MIT Spam Conference eyes filters to destroy spam business model [InfoWorld: Top News]

In tests Yerazunis performed, MailFilter was 99.915 percent accurate in identifying spam. "I'm only 99.84 percent accurate at identifying spam, so this is much more accurate than I am," Yerazunis quipped.

I am happy for the filter freaks, but its mostly an academic exercise.

Filters are not a solution because they result in false positives, solution techniques are open, spammers adapt, and volume grows.  Legal solutions generally wont work because of juristiction, aside from putting bounty hunters on the task.  Its fundamentally an economic problem that requires an economic solution -- raising the cost of sending email, but having cost vary according to trust networks.  One day we might actually be able to accept these costs.

Spam Conference Trip Report

Yesterday was the first ever Spam Conference and it was held at MIT.  First off I have to give huge kudos to Paul Graham the organizer.  In a little over a month he put together a truly outstanding conference --- and the interest level was astonishing.  They expected "50 to 60" and instead 560 signed up.  And since the room was basically full and it held 566 by number of seats, I'd say roughly 520 to 540 actually made it.  And don't think that these were all local MIT geeks either; I sat next to a researcher from IBM Zurich, had lunch with people from Cloudmark (San Francisco), met the founder of pobox (Philadelphia), spoke with Tony Bowden at length (England) and others.  Presentations came from BrightMail, Popfile, Microsoft France, MIT, Mitsubishi Electric Research Lab, ShopIP, MessageLabs and others.

Most of the discussions focused on "Naive Statistical Bayesian Classifiers" such as iFile and Popfile (although many other systems were represented including other types).  If I can find the papers online I'll post links to them.  Otherwise Google for them from the info at the Spam Conference page.  More on the conference later or tomorrow.