- cross-posted to:
- tech@piefed.world
- cross-posted to:
- tech@piefed.world
You must log in or register to comment.
AI companies cannot be allowed to hide behind “it wasn’t me!” defences.
SHAGGY
Ha ha, I’m wishing I’d said that myself
And yet watch as they do exactly that for the next several decades.
Why should OpenAI be held responsible for this particular incident? Because their LLM recommended the user to contact a suicide hotline when he expressed his suicidal thoughts? Because it correctly answered his questions and confirmed that the safety of his shotgun was off? Don’t get me wrong, I like having companies held accountable as much as the next guy, but why don’t you ask the gun production companies to be held responsible as well? Hell, we might as well be asking for his parents to be held accountable for their offspring’s actions.
Yes, LLM’s can, with or without external manipulation, provide information about sensitive subjects, which can then be consumed by vulnerable individuals who may act in a harmful way. That individual would have acted and reached the same outcome not matter if the LLM would have existed or not, because there are multiple options to using an LMM to gather information.
What are the realistic expectations of OpenAI to implement to prevent these things, that doesn’t automatically mean stricter restrictions to the user data privacy and further strengthen the control of citizens (thanks to the flourishing relationship between OpenAI and the American government)?
I thought it was interesting how the article said that if ChatGPT was a human it would have been charged with murder.
Obviously I don’t think ChatGPT is a human but it does raise questions about the kind of laws regarding LLM and the companies behind them.
Like at what point do AI companies have a responsibility to step in in these scenarios?
They don’t. It’s not human.
Company is still ran by people though
I have to wonder if these bots have special protections in place for their CEOs. Like you want to kill your teacher? Sure whatever. Here’s an ad. Want to kill the head of the company that runs that bot? Hmmmmm
As someone who contends regularly with suicidality, I have to admit the 988 service leaves something to be desired. Conversations can be distilled to essentially this:
Me: Everything that I was raised to believe about what my society is trying to be turned out to be a lie. We’re not trying to be a plurality that believes in social equality. Most voters in the US are racist and xenophobic. They’ve raised a regime paramilitary to route out undesirables and lock them into torture prisons, or send them into destabilized nations where they’re either imprisoned or trafficked into slave labor. And they coming for me. I’m thinking, I’d rather die quickly and painlessly than from being tortured to death for months in some concentration camp. I need an exit plan.
988: Have you tried distracting?
I’m not an imminent danger to myself or others, and am extremely unlikely to go amuck, but the way society is crunching the common American working class, it does not surprise me at all if those with the inclination to radicalize are doing so in record numbers.
I recently listened to something very similar. I guess there’s been several cases by now. Not to speak of all the suicides.
What A-so-called-I providers are doing is the equivalent of mass producing cars that haven’t been tested for anything, then sell them for $1 each just to get more valuable data from how people drive them.
It’s pretty interesting article. I actually dropped my Chad GPT subscription a few years ago and switched a locally hosting my own Ollama server for like privacy concerns.
I’m still going to continue to use my locally hosted large language models but I don’t know, its interesting to know that chat gpt seems to not care about user data except for training or making money.
Ollama is better than GPT, but worse than basically every other local solution (from a performance perspective and an ethical perspective). I was annoyed when I found out, but made the switch to something llama based.
