And Bayes' Rule, equally unambiguous, says that an email containing both words would, in the (unlikely) absence of any other evidence, have a 99.97% chance of being a spam.
This statement is wrong -- it would only be correct if a spam containing sex and containing sexy were independant which seems definately wrong.
If a person sent him an email containing both the words sex and sexy (and perhaps a few other related words), which seems very possible, the probability of being a spam will go way too high and it will be very hard for the system to classify it as non spam. This might seem inevitable, but it doesn't have to be the case.
Of course generating the entire joint distribution over all possible words is impossible, but there are very good approximations, for example he could use a Bayes net.
I don't think this would work. At some point the message goes in unencrypted. This would only make the FBI's work slightly more difficult and with all the money they seem to have to waste I doub't it would do anything to stop them.
Slashdot Mirror
For those of you that enjoy interesting and different games, you should check this out. It's a 3d arcade version of the ancient chinese game Go.
Too bad Art Bell is no longer on the air. That's the only thing on the radio that would be worth recording.
skipping commercials
Damn that network pong was top notch. Holler.
And Bayes' Rule, equally unambiguous, says that an email containing both words would, in the (unlikely) absence of any other evidence, have a 99.97% chance of being a spam.
This statement is wrong -- it would only be correct if a spam containing sex and containing sexy were independant which seems definately wrong.
If a person sent him an email containing both the words sex and sexy (and perhaps a few other related words), which seems very possible, the probability of being a spam will go way too high and it will be very hard for the system to classify it as non spam. This might seem inevitable, but it doesn't have to be the case.
Of course generating the entire joint distribution over all possible words is impossible, but there are very good approximations, for example he could use a Bayes net.
I don't think this would work. At some point the message goes in unencrypted. This would only make the FBI's work slightly more difficult and with all the money they seem to have to waste I doub't it would do anything to stop them.
If microsoft put this feature in IE it would disable their entire site.