Maybe the real lesson is it was trained on human-written comments in discussion fora, so it perfectly mimics the average lack of point, weak arguments, rambling

Maybe the real lesson is it was trained on human-written comments in discussion fora, so it perfectly mimics the average lack of point, weak arguments, rambling and incoherence in fora? It would be interesting to see if the output has a similar quality when trained only on highly regarded texts.