Hacker News Reader

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

Maybe but having it search first to load the context with relevant information sure gets better results

Included in the training corpus doesn’t mean perfect or even partial recall.