agree, prompting a token predictor like you’re talking to a person is counterproductive and I too wish it would stop

the models consistently spew slop when one does it, I have no idea where positive reinforcement for that behavior is coming from