A core issue here is that there isn't a black and white on a subject like this. Yes, it is wrong. Yes they have a responsibility. But at the same time taking that to an extreme leads to heavy censorship. So what is a practical middle ground? Is there something like a 'sexualized validation suite' that could be an industry standard for testing if an LLM needs additional training? If there were then victims could potentially claim negligence if they aren't using best practices and they were harmed because of it right? Are there missing social or legal mechanisms to deal with misuse? One thing I think is missing is a '911' for cyber offenses like this. If someone breaks into my house I can call 911, if someone creates revenge porn who do I call? I don't think there is a simple answer here, but constructive suggestions, suggestions that do balance free speech and being a responsible service provider would be helpful. 'They are wrong' doesn't actually lead to change.
Looks like this hit a nerve. Any comments on the practical solutions though? The comment wasn't advocating that they should make CSAM or that they shouldn't face repercussions for enabling it, at least I don't think it reads that way. I honestly think that a core issue here is we are missing practical fixes. Things that make it easier for victims to get relief and things that make it clear that a provider is being irresponsible so that they can face civil or criminal penalties. If there aren't solid industry standards then how can you claim they aren't implementing best practices to hold them accountable? If victims don't have effective means of relief then how will we find and stop this? I'd love to hear actual concrete actions that the industry can put in place. 'Just tell them to stop' doesn't create a framework that leads to change.
The reason it hit a nerve is that you're just being extraordinarily credulous about xAI's lies. There are solid industry standards, and we can just tell them to stop; we know this because Grok has a number of competitors which don't generate CSAM. Indeed, they've already implemented the industry standards which prevent CSAM generation; they just added a flag called "spicy mode" to turn them off, because those standards also prevent the generation of pornographic images.
Trust me, I believe nothing positive about xAI. Various players doing similar things and an actual published standard or a standards body are totally different things. The industry is really young. Like a couple years young a this point. There really aren't well developed standards and best practices. Moments like this are opportunities to actually develop them and use them, or at least start the process. Do you have a recognized standard you can point to for this? When it comes to car safety there are lots of recognized standards. Same with medical safety, etc etc. Is there anything like that actually in the LLM world?
