Allowing them to do analysis or summary of anything that isn't already in a Wikipedia entry is radically dangerous because of how good they are at hallucinating and lying (and we can't detect that at scale). It is like an automated "yes man". This is the most dangerous bubble in living memory and the cost of it popping will be in the failure of any industry that is foolish enough to rely on them at scale.
And to attack that study in particular, "help people who couldn't write well write better" is complicated, increasing accessibility is good, but there is a BIG difference in an increase in writing quality versus an increase in productivity AND correctness.