I think it would be a lot of effort to keep changing the paper until you get the perfect abstract. It would be easier to train different models or do random sa

I think it would be a lot of effort to keep changing the paper until you get the perfect abstract. It would be easier to train different models or do random sampling from the predicted distribution.