4 ms·
In my case, the spam that manages to evade Gmail's filters contain an image as the body instead of text. Fun fact: Back in 2002 ycombinator founder Paul Graham
by flowerlad 5y ago
In my case, the spam that manages to evade Gmail's filters contain an image as the body instead of text.
Fun fact: Back in 2002 ycombinator founder Paul Graham wrote an article on spam filtering. (See http://www.paulgraham.com/spam.html http://www.paulgraham.com/spam.html ). I emailed him that his method can be defeated by sending an image of the text, as opposed to the text itself. PG replied and pointed me to this FAQ: http://www.paulgraham.com/spamfaq.html http://www.paulgraham.com/spamfaq.html
In the FAQ there is an entry named "What if spammers sent their messages as images?" The answer indicates that is not going to be an issue, because there's still plenty of signals to go by.
Guess PG was wrong!
- pensatoio 5y agoExactly the same issue here.
- Tagbert 5y agoBut I never see images in most emails. They are not loaded unless I decide I want to see them. Those emails are just empty space. If there is no unsubscribe link, they get spam tagged.
- alex_smart 5y ago>Guess PG was wrong! I mean, he wasn't wrong. He was describing the state of spam at the time. The way I see it, you were trying to corner him into giving the answer you wanted, and he was trying to corner you into asking the question that would force him to give the answer you wanted. All machine learning models make intrinsic assumptions about their input distribution (no free lunch theorem and all that). If all your objection was that the model will fail to work if the assumptions it makes about the input distribution will fail to hold, that is a criticism for the entire field of machine learning, not this one spam-filtering product.