Comment by ori_b

It's worse than that. LLMs have been tuned carefully to mostly produce output that will be inoffensive in a corporate environment. This isn't an unbiased sampling.