I mean, having a curated dataset of the works and posts of the top 200 coders in the world (at least the public ones) is not very difficult. I’m sure these articles like the one in OP will be very easy to mark as “high value training data”. I think you’re letting your bias blind you