While there's a lot of reason to be dubious that this can scale to deal with all of Twitter's troll problems, bear in mind that there are people who are literal

While there's a lot of reason to be dubious that this can scale to deal with all of Twitter's troll problems, bear in mind that there are people who are literally using Twitter to make death threats (I don't think it makes sense to call people who do that "trolls", as they're far worse, but I guess that's a losing battle). A system designed to catch "I'm gonna rape and kill you" can be a lot dumber and closer to automated than something designed to moderate all comments (I don't think twitter should be moderating comments. I do think they should be trying to address threats).