Публикация

I saw that the models struggled with negation so I would not use they for summaries at all. Somebody explained that that isn't strange as text statistically wise the sentences 'that is true' and 'that is not true' are pretty similar.