I don't think there is much emotion in said post. I am making specific assertions.
to your point:
> Non-sexualized bikinis are sold everywhere
Correct! the key logical modifier is Non sexual. Also you'll note that a lot of clothing companies do not show images of children in swimwear. Partly that's down to what I imagine you would term puritanism, but also legal counsel. The definition of a CSAM is loose enough (in some jurisdictions) to cover swimwear, depending on context. That context is challenging. A parent looking for clothes that will fit/suit their child is clearly not sexualised (corner cases exist, as I said context) Someone else who is using if for sexual purposes is not.
and because like GPL3 CSAM is infectious, the tariff for both company and end user is rather high for making, storing, transmitting and downloading those images. If someone is convicted of collecting those images, and using them for a sexual purpose, then images that were created that were not-CSAM suddenly become CSAM, and legally toxic to posses. (context does come in here.)
> Your link clearly states they are being moderated in the comments
Which tells us that there is a lot of work on guardrails right? its a choice by xai to allow users to do this. (mainly the app is hamstrung so that you have to pay for the spicy mode.) Whether its done by an ML model or not is irrelevant. Knowingly allowing CSAM generation and transmission is illegal. if you or I were to host an ML model that allows user to do the same thing, we would be in jail. There is a reason why other companies are not doing this.
The law must be applied equally, regardless of wealth, or power. I think that is my main objection to all of this. its clearly CSAM, and anyone other than musk doing this would have been censured by now. All of this justification is because of who it is doing this, rather than what is being done. We can bike shed all we want about is it actually really CSAM, which negates the entire point of this, which is its clearly breaking the law.
> The GPT models wouldn't know what is sexualised.
ML Classification is really rather good now. Instagram's unsupervised categorisation model is really rather effective at working out context of an image or video (ie differentiation of clothes, and context of those clothes.)
> please don't put words in my mouth
I have not done this, I am asserting that the bar for justifying this kind of content, which is clearly illegal and easily prevented (ie a picture of a minor and "generate an image of her in sexy clothes") is very high.
Now you could argue that I'm implying that you have something to hide. I am actually curious as to your motives for justifying the knowing creation of sexualised images of minors. You've made a weak argument that there are legitimate purposes. You then argue that its a slippery slope.
Is your fear that this brings justifies an age gated internet? censorship? What is the price that you think is worth paying?