[Houshalter]

These general QA datasets wouldn't work, because their system can only do word analogies. And some similar tasks like antonyms. It can't understand a sentence or anything.