On Wed, Jan 08, 2003 at 05:22:41PM -0500, rODbegbie wrote: # An advertising bulletin that I get as a former customer of a company # might be considered "ham" by me, but "spam" by someone else.
I hear this a lot, but in practice it's not really a strong argument. Based on the SpamNet system currently in place, it's possible for us to measure just how much people disagree, and look at what the disagree on. Very few disagreements involve SPAM, most of them are about real mailing lists. Not saying it doesn't happen, just trying to put the statement into perspective -- for the most part we see everyone agree on 98%+ of what is reported, sans mailing lists (whose reports seem more a function of laziness than anything else). # You could have the most perfectly designed hashing and trust # algorithms in the world, but there'd still be false positives and # negatives. Well, sure, anything non-human will for sure have false positives and false negatives. (Humans too, but shh, don't tell Brightmail! ;-) ) The hope here is that a system using direct correlation of distributed human input (which is harder to fool) based on reputation can avoid certain mistakes that pure machine-based systems would otherwise fall victim to. It may itself be prone to errors as well, but this game is particularly different and thus we arguably stand a better chance (though by how much is debatable) at getting it right -- people have been defeating rules-based systems for years, but reputation systems and advanced fingerprinting algorithms are more sophisticated/less understood in general. --jordan
msg00520/pgp00000.pgp
Description: PGP signature
