If you couple this with voice input, I can very easily go through a few posts and give detailed feedback on each in less than a minute; it means no longer relying on inferences about the categories or collaborative filtering to guess at user preference.

Recommendation engines based on "show more / show less like this" are already feeling outdated. With LLMs we're starting to enter an era where we can just tell our computer what we like and why we like it and the computer will actually understand.