Yea perhaps more generally, the LLMs might model the average HN comment (sample mean) quite well. But, they fail to model the sample variance at all.