[paulkon]

How do humans with very little exposure to grotesque violence or extreme content universally label such content so well? This is not graduate level data that needs labeling.

What is missing in an AI model for it to intuitively understand what content is extreme from very few labeled sample in training?

[hexaga]

A finely tuned set of heuristic triggers for fear, horror, disgust, etc. You might as well ask why pain is so painful.