Comment by Aeolun

Would it be hard to create such a training dataset? Seems like you’d just need a lot of people to say a bunch of stuff for you?