From their FAQs:
Q: How are you going to manage to recognize stupidity programmatically?
A: Pretty much the same way you can programmatically recognize spam, we'll look for things that characterize stupidity and assign particular tokens different weights based on how often they occur in hand-picked examples of idiotic comments.
Q: Isn't filtering stupidity elitist?
A: Yes. Yes, it is. That's sort of the whole point.
Q: Aren't you just trying to eliminate comments and discourse that you consider to be stupid?
A: As much as that might be nice, no. The StupidFilter does not understand, in a meaningful sense, the text that it parses, and our graders select comments that are formally stupid -- that is, their diction, not their content, marks them as stupid. It is not our intent to eliminate debate or disagreement, but rather to programmatically enforce a certain quality of expression. Put another way: The StupidFilter will cheerfully approve an eloquent, properly-capitalized defense of mandatory, state-subsidized rocket-launcher ownership for all schoolchildren.