Comment by dalmo3

Is it not safe to assume that all* publicly available prior work is in the training data?

Then you could just prompt it to propose options with pros and cons etc.

* Bar extremely new stuff from after the cutoff