If someone published a book advising people to take drugs, would people be filing lawsuits? No—we would agree that people are allowed to write whatever they want, even if what they say is terrible, right?

I really think these criticisms are misguided. I realize an LLM is not a person—but it does still represent speech, and certainly, any guardrails put in place would themselves be human-authored speech. There are all sorts of social norms which I personally believe, but which I don’t want AI companies to be enforcing on everyone.

Imagine if ChatGPT had launched 50 years ago, before LGBT acceptance was mainstream. If ChatGPT had told users “it’s okay that you’re a boy and you like other boys, pursue your instincts”, people would have been screaming from the hills that ChatGPT was turning their children gay. They might have tried filing lawsuits. Do we really want to allow that?

> If someone published a book advising people to take drugs, would people be filing lawsuits? No—we would agree that people are allowed to write whatever they want, even if what they say is terrible, right?

That's not the situation here. The more accurate case would be:

> If someone without a medical license provided blatantly incorrect medical advice with respect to safe medication usage to an individual via a direct one-on-one discussion, would people be filing lawsuits?

And the answer is yes. You can be wrong and you can say incorrect things. What you can't do is provide medical advice unless you are a licensed medical professional. You can still speak about medical topics but you have to disclaim your lack of licensure. You have to make it clear that you are not providing medical advice.

If this was a person doing this it'd be a crime, clear as day. It's called "practicing medicine without a license" and in the US it is a criminal offense in all 50 states, Washington DC, and all 5 inhabited territories. Whether it is a misdemeanor or a felony is dependent on the jurisdiction and the case but it's a crime everywhere in the US.

The issue was not "you should take drugs."

The thing that killed this person was being advised to take xanax while having a lot of kratom and alcohol in their system. And yeah, if you published a book telling people that xanax is a great treatment for alcohol induced nausea and people died following this advice you should go to prison.

There's a bit of a difference between "enforcing social norms" and telling a user to ingest prescription drugs to combat nausea from the other drugs that it told the user to take.

Yes, you should be able to write a book with this same information. No, you should not be able to release software that instructs its users to harm themselves. LLMs aren't people, and you shouldn't anthropomorphize human rights onto them.

Agreed - people should learn the ChatGPT does not give good advice, but the question is did OpenAI advertise ChatGPT as a good and reliable source of information on health ???.

There's soft guardrails for "reputable" content. A publishing house has to buy it, stores have to agree to distribute it, and if people are upset they can raise a stink and get the book pulled.

Technically, people can write whatever they want, but practically you can't walk into a bookstore and read whatever you want.