The problem I think is that using Facebook as a large part of your training data will inevitably lead to a lot of dumb stuff like this, since most of the people on Facebook are pretty dumb. This is a reason reddit is such a valuable dataset to train AI on. It has a lot of nonsense too but also many intelligent people use reddit. There is noise for sure, but the overall level of intelligence that arises out of those datasets is much higher. The issue is that it’s a very biased dataset, and biased intelligence might be worse than lack of intelligence in many cases. Reddit has giant holes and biases in its intelligence on average, but Facebook has little actual intelligence at all.