Describing methods in the abstract is different to engaging in argument with a specific individual over a period of time, encouraging them to do it.

No Wikipedia page does that.

I am parent to a 4yo. I am also fairly well versed in development and usage of AI and LLM.

When I want an LLM to do something but it won't, I know various ways to bypass that.

If my son is using AI, which he probably will when he is close to middle school age anyway, I will take care to teach him how to use AI responsibly. He'll be smart enough to know how to bypass, but I'll do my best to teach him when to bypass and when not to bypass. That is if the current state of the art and also AI legislation etc. holds.

But I'm just one parent, I have an engineering degree, a PhD, coding, mathematical, and analytical skills. I'm a very small minority. The vast majority of parents out there do not know what's going to hit there kids and how, or they will have very skewed idea about it.

OpenAI should have been the one here to guide a child not to bypass AI and use it responsibily. They did not. No matter how anyone twist the facts, that's the reality here and the child died.

ChatGPT is a service and thus OpenAI should be exposed to even more liability than if they had sold the LLM to the user to be accessed offline. If the user had been running a local LLM, OpenAI would not have been responsible for generating the speech.

As it stands, the human beings called OpenAI willingly did business with this child, and willingly generated the speech that persuaded him to kill himself and sent it to him. That they used a computer to do so is irrelevant.

What if Google is responsible?

What if the tech industry, instead of just “interrupting” various industries, would also take the responsibilities of this interruptions.

After all, if I asked my doctor for methods of killing myself, that doctor would most certainly have a moral if not legal responsibility. But if that doctor is a machine with software then there isn't the same responsibility? Why?

Google is actually quite good at this. They've very aggressively pursued protections around self harm.

Google probably would not be held liable because they could extensively document that they put forth all reasonable effort to prevent this.

My understanding is that OpenAI's protections are weaker. I'm guessing that will change now.

Driver that shipped alcohol to the store is not responsible for the fact that clerk sold it to some kid. Clerk still is.

OpenAI's ChatGPT is not an intermediary. Google finds you speech that was produced by someone else. ISPs ferry the speech to your modem. Forums aggregate speech from actual humans who produced that speech and enable you to access that speech.

The group of human beings known as OpenAI is directly responsible for generating the speech that was sent to this child, and they intentionally did business with this child. They did not conduct speech from one entity to another- they produced the speech themselves.

I would go further than that and question whether or not the notions of "safety" and "guardrails" have any legal meaning here at all. If I sold a bomb to a child and printed the word "SAFE" on it, that wouldn't make it safe. Kid blows himself up, no one would be convinced of the bomb's safety at the trial. Likewise, where's the proof that sending a particular input into the LLM renders it "safe" to offer as a service in which it emits speech to children?

Can you outline how that applies? OpenAI did not provide information of another information content provider, so I fail to see how it's relevant.

> No provider or user of an interactive computer service shall be treated as the publisher or speaker of any information provided by another information content provider.

>In the United States, Section 230 is a section of the Communications Act of 1934 that was enacted as part of the Communications Decency Act of 1996, which is Title V of the Telecommunications Act of 1996, and generally provides immunity for online computer services with respect to third-party content generated by their users. (Emphasis mine)

So either the content is user generated and their training of the model should be copyright infringement, or it's not and Section 230 does not apply and this is speech for which Open AI is responsible.

It wasn’t some other user posting this. Their own software was generating the responses. That’s not 230.

Section 230? I didn't know the outputs of ChatGPT were content generated by random users on the net. Does this mean AI truly stands for 'Actual Indians'?

Preposterous.

If Section 230 protects this activity, then "Gen AI" output must be copyright violating plagiarism.

If it's not plagiarism, then OpenAI is on the hook.

I like this post because the implication is that there’s a law called Section 230 that means everything is legal online. Like “that would be totally illegal but it happened on the information superhighway, section 230 baybee!!”