Do you generate the headline from the original headline, or do you desensationalize the body of the article and then ask OpenAI to generate a headline based off of the new article body?
It seems like the latter might be better, since it ensures the headline actually matches the article below, as opposed to relying on a likely-clickbaity title that would no longer match the desensationalized article body.
Probably worth testing both ways to see what the results look like!
Also, in a couple of cases the desensationalising has removed (useful) information. E.g. in https://www.boringreport.org/app/all/645d396cf7c90670355a6c7... the headline is "Supreme Court Decisions in Two Cases" missing any of the stuff about corruption in the original headline "Supreme Court sides with ex-Cuomo aide and Buffalo developer in disputes over corruption convictions".
(also, that click jacking to prevent copy/paste is a PITA)
It seems like the latter might be better, since it ensures the headline actually matches the article below, as opposed to relying on a likely-clickbaity title that would no longer match the desensationalized article body.
Probably worth testing both ways to see what the results look like!