People are probably thinking data more complex than normal distributions (though I'm also not sure if GPT-4 is the best method for that)