2. Lots of users to label the email as spam or not spam.
3. Tons of code to identify classifiable features such as clickable links, the ASN from which the message originated, and so many other things.
4. Some kind of classifier that pulls all the above into a decision about each message. And make sure it’s really fast and doesn’t require gargantuan resources either, because you’re trying to provide cheap mailboxes.
Google has enormous advantages in solving for spam filtering. It has over a billion active users, enormous computing resources, and top AI talent. Not to mention legions of talented general programmers to build all the feature extraction stuff. Competing with Google on spam filtering is an unwinnable game IMHO.
All I know is what I can see from the X-Spam headers they add with that matched rules, which are SA's. FastMail has been around for ages, and I guess it's "good enough" (or at least, not "bad enough" to replace). It's not so easy to replace a tested (though imperfect) solution like SA with some new AI stuff, even of it would be better; it's all about ROI.
Also, not sure if Howard is still involved to FastMail? They sold to Opera in 2010, although they've been independent again since 2013, but it looks like the 2010 sale was the end of Howard's involvement?
1. Lots of email that you can categorize.
2. Lots of users to label the email as spam or not spam.
3. Tons of code to identify classifiable features such as clickable links, the ASN from which the message originated, and so many other things.
4. Some kind of classifier that pulls all the above into a decision about each message. And make sure it’s really fast and doesn’t require gargantuan resources either, because you’re trying to provide cheap mailboxes.
Google has enormous advantages in solving for spam filtering. It has over a billion active users, enormous computing resources, and top AI talent. Not to mention legions of talented general programmers to build all the feature extraction stuff. Competing with Google on spam filtering is an unwinnable game IMHO.