Hacker News Reader

People use down and upvotes based on what they like/dislike.

Thus, most top comments are jokes.

Anything informed but controversial will be buried down deep and can be only found by sorting for..controversial.

GPT-2's dataset used every reddit post with more than 3 net votes as training data, IIRC. Seems likely they've iterated on that approach since.

Voting there isn’t real, many subreddits are simply echo chambers with the hive mind downvoting because you said something different

Perhaps that could be normalized for? And if you consider multiple subreddits you might recover some diversity of opinions?

There's no content to balance it out with. Stepping outside of the popular opinion and getting immediately downvoted so nobody sees it, you might as well not have commented at all. Eventually all dissenters stop commenting.

But you can go to non-mainstream subs where there's a different voting population and that doesn't happen (well, it does, but the "popular opinion" can be different within the sub).