You're right bias is hard. Since most scientists are men, I guess those pictures have the bias correct. But if you asked it 100 times, would the unbiased result

You're right bias is hard. Since most scientists are men, I guess those pictures have the bias correct. But if you asked it 100 times, would the unbiased result be 100 men because each picture is of a typical scientist or should it be the same ratio as real life because it's supposed to act like it's taking a random sample each time? All pictures of people LLMs produce are deeply biased. He's also posing like a stock photo model instead of picking his nose or whatever random activity a real scientist might be doing if captured in an unbiased photo. Poses are something it's pretty biased on in general, as well as dress style, facial expression, attractiveness, etc. And it has to be, because that's all part of integrating broad common sense knowledge and is what we want when we ask it for pictures. We're not usually looking for a random or average example but something that matches our cultural-artistic expectations.