There are many examples of people testing LLMs on which is worse, extinguishing all of humanity or making a racist comment in a place where no one will hear it. The LLMs keep choosing to extinguish humanity. Just something to consider when using "but it could say something racist" as justification for filtering. Is racism so evil that it's better to kill every human to avoid it ever happening? Maybe the only option in that case is to prohibit the existence of LLMs. We already have corporations in the role of powerful sociopaths preying on society, do we need LLMs looking to kill us all to satisfy its goals too?