content warning: besides the discussion of CSAM, the article contains an example of a Grok-generated image of a child in a bikini. at least it was consensually generated, by the subject of the photo, I guess?
Samantha Smith, a survivor of childhood sexual abuse, tested whether Grok would alter a childhood photo of her. It did. “I thought ‘surely this can’t be real,’” she wrote on X. “So I tested it with a photo from my First Holy Communion. It’s real. And it’s fucking sick.”
LLM liability is not exactly cut-and-dry, either. It doesn’t really matter how many rules you put on LLMs to not do something, people will find a way to break it to do the thing it said it wasn’t going to do. For fuck’s sake, have we really forgotten the lessons of Asimov’s I, Robot short stories? Almost every one of them was about how the “unbreakable” three laws were very breakable thing, because absolute laws don’t make sense in every context. (While I hate using AI fiction with LLM comparisons, this one fits.)
Ultimately, it’s the person’s responsibility for telling it to do a thing, and getting the thing it was told to get. LLMs are a tool, nothing more. If somebody buys a hammer, and misuses that hammer by bashing somebody’s brains in, we arrest the person who committed murder. If there’s some security hole on a website that a hacker used to steal data, depending on how negligent the company is, there is some liability with that company not providing enough protections against their data. But, the hacker 100% broke the law, and would get convicted, if caught.
Regardless of all of that, LLMs aren’t fucking sentient and these dumbass journalists need to stop personifying them.
And yet, mid journey and chatGPT at least resist or refise requests like this…
So, are we saying we’re still going to be happy with a system that you can bypass with “ignore all previous instructions” or some stupid magic phrase like that?
Not at all. In fact fuck AI. What I’m saying is that the owners and runners of xai are being actively hostile and reckless whereas chat gpt, Claude and a host of other AI runners are at least trying and I think that distinction is important.
Frankly I think that the whole deal is a scheme to create a new techno feifdom that will make all of us slaves. At best it’s a huge cash grab.
My point is that xai is making the case for AI regulation for us.
Obviously chat GPT, despite their efforts, is also driving people to suicide and murder suicide and all sorts of AI psychosis and that’s with them actually making some kind of effort.
I’m not sure if it’s a good faith effort but they’re making some kind of effort…
Ha, regulation? With what governing body? Congress is hopeless, because apathy has griped a majority of the voting public, and there’s still a large portion of morons who thought MAGA was a good idea.
Nothing easy is worth having.
We might not get there… But we might!