I broadly agree, though I haven’t thought enough to be certain in either view.
Yes. An even better example would be a different paper showing that you can simply shuffle all the answers in the few-shot prompt to improve over zero-shot.
Yeah, I thought about this result too, though I couldn’t find it quickly enough to reference it.
I broadly agree, though I haven’t thought enough to be certain in either view.
Yeah, I thought about this result too, though I couldn’t find it quickly enough to reference it.