Can you provide a link with evidence of that? I haven't seen that reported.
I'd also note in advance there is a big difference in someone figuring out how to jailbreak Gemini or OpenAI, and then the companies responding swiftly to fix that, than what has been reported with Grok where it was basically wide open to create those images.
Grok has never been "wide open" to undress people. Anyone reporting that is being extremely duplicitous. Any image-to-image on Grok has stringent NSFW filters for exactly this scenario. People have worked out jailbreaks and those get dealt with.
Gemini and ChatGPT conversations are private, not public. A big part of the controversy over Grok is that it's happening in public on Twitter, often as direct replies to the user who's picture is being manipulated.
This is such a weird issue for me, who is blind. Did Grok undress people, or did Grok show extrapolated images of what people might look like undressed? The "undressed people" framing makes it sound like people physically had their clothes removed. Obviously this did not happen.
But, like.
If I have like ... a mole somewhere under my clothes, Grok cannot know about that right? People will know what they themselves look like naked?
Someone looking at Grok's output learns literally nothing about what the actual person looks like naked, right?
Kinda sounds like somebody should just make something that creates this for every picture ever? Then everybody has a defense -- "fake nudes!"
and the pictures are meaningless?
> This is such a weird issue for me, who is blind.
I'm not sure what your mental model is for someone's visual likeness.
I'd propose a blind-inclusive analogy of what is happening on Twitter is anyone can create a realistic sexdoll with the same face and body proportions as any user online.
Doesn't that feel gross, even if the sexdoll's genitalia wouldn't match the real person's?
I interpreted your last sentence as asserting it was no big deal (ie not gross) because it was all fake, but fair enough if you didn't mean it that way.
But to your main point: if you agree it's gross, do you not agree it is a violation of _something_? What is that thing if not privacy?
You may disagree, but 95% of people in the real world understand what "undressed" means in this context and see it as a gross invasion of privacy.
I knew when this issue hit the fan that you'd get hordes of overly-literal engineer types arguing that the person wasn't actually violated, or that "how is this any different from someone drawing a hyper-realistic picture of someone naked?" I can actually even (well, somewhat anyway) sort of understand this viewpoint. But if you want to die on this hill, you will, most people in the real world would condemn and ostracize you for this viewpoint.
This is the sort of thing that is technically correct, but misses the emotional aspect that people want to be able to control their public perception. Of course people could (and did) do this with older tools or by hand. It doesn't matter to them. And since Elon/X are the villain du jour, it's a good lever to punish them.
Use the mole example as referring to any physical characteristic hidden by clothing that people want to remain hidden. It's an example to demonstrate that the AI is not "undressing" anybody. It is filling in an extrapolation of pixels which have no clear relationship to the underlying reality. If you have a hidden tattoo, that tattoo is still not visible.
This gets fuzzy because literally everything is correlated -- it may be possible to infer that you are the type of person who might have a tattoo there? But grok doesn't have access to anything that hasn't already been shared. Grok is not undressing anybody, the people using it to generate these images aren't undressing anybody, they are generating fake nudes which have no more relationship to reality than someone taking your public blog posts and then attempting to write a post in your voice.
sure, but if I make a fake picture of someone having sex with a horse and someone else confirms "my gosh, that's really them! I recognize that mole" then I suppose it is the same damage.
At any rate where some of this stuff is concerned, fake CSAM for example, it doesn't matter that it is "fake" as fakes of the material is also against the law in some places at least.
if the problem is just the use of the word "undressing" I suppose the usage of the word is completely analogical, as nobody expects that Grok is actually going out and undressing anyone as the robots are not ready for that task yet.
It is reductio ad absurdum applied correctly. I made a completely absurd proposal, because the response is obvious - any consideration of "fake porn being fake" will stop mattering super fast once actual faces are put in the picture.
Invoking the image of someone's loved ones and then imposing a vicious scene on top of it is a way of inflicting pain on that person. That kind of violation is certainly a personal attack. Actually, it's even more of a personal attack than attacking the individual in the narrow sense.
You can't do that here and we ban such accounts, so please refrain.
p.s. it's also a text-based way of doing the very thing you're inveighing against, so it's not in your interest to argue that way.
Yeah the reason Google and OpenAI etc are silent is because their services do the same but they "aren't the bad guys" so if they shut up the crisis will pass.
This of course implies that the crisis itself and persecution of Musk/Grok is politically motivated, or just based on stupidity.
The same capabilities might be present in many available models, but I do think that the public/social aspect in usage is quite different— people can’t come into my Google account and save nudified versions of my family photos directly to my Google drive, but X generated a lot of attention because the users are directly replying or quoting other users and @ing them with the modified photos.
Solving autonomy is the hardest technical challenge right now, way harder than creating a frontier LLM AI. Tesla has basically solved it. So to argue sarcastically about Tesla is in my opinion being ignorant. I get the Elon hate, but he is usually correct in his predictions, albeit late
Waymo has solved it. Tesla and BYD probably will solve it. And then everyone else will solve it for the same reason everyone likes having car factories: jobs and tanks.
I say this as someone who was in a Waymo and used Tesla’s latest FSD less than a month ago. One of them still fails spectacularly ungracefully. In the other I can take a nap.
Waymo has not solved generalized autonomy. They need training wheels and HD maps for every city they expand into. You can't just put a Waymo into any city and expect it to drive autonomously.
The question I ask every person pitching FSD is whether they would let the FSD take their kid to soccer/bball… practice - still waiting for first yes answer
We have a pretty good idea of what products Apple is likely to release next year.
We no longer have an actual product roadmap of revenue viable product releases from Elon and Tesla. It’s just irrational market exuberance but all the products are being abandoned.
That isn’t really sustainable.
I drive a Y. I’ve said before that Elon was peak Elon when they released the 3/Y and he sent the Roadster into space. Everything since has been a grift of diminishing returns, in terms of reality and actual hardline income.
Grok and xAi are a thing but they mostly cost money right now. I kinda wonder if he isn’t irrationally energetic enough to get something interesting out of Macrohard but he can no longer iterate product releases to save his life.
My other long bet is that Google will win a big chunk of AI (because TPUs and frontier and other income and enterprise sales) and Elon will mostly succeed in outspending OpenAI and plundering their chances (save the magical odds of their Jony Ive bet delivering some kind of post-moat income autonomy). xAI and OpenAI will sink into the deep together.
What exactly has Tesla solved, automation-wise? Not sure what you are specifically referring to? I am being everything except ignorant, I look at things with my own eyes and do not fall for car-salesman tricks. Perhaps I gave him the benefit of the doubt initially but after decade+ of overpromising and underdelivering (underdelivering might be the understatement of the century) forgive me if I do not believe what a car salesman is pitching. I do not hate Elon at all, actually think he's one of the greatest visionaries of our time and probably the greatest salesman in the history of mankind.
True, we need to see the details of how they compute all this data. I do remember reading that they include crash data up to like a minute after FSD has been disengaged so these types of crashes should be included.
Nevertheless, the latest version of FSD I can certainly believe is seven times less likely to get into an accident than the average driver. I experienced it daily.
When the stakes are that high, words such as hypocrite don’t fit very well. It’s all game theory. It bought us Pax Americana for a while. Not universally loved but a time may (God forbid) come when it will be universally missed.
It also makes this horrible kind of sense that Elon would see them both as admirable, this idea that you're the only person who matters. Ordinary people exist only for you to exploit them, and have no intrinsic worth.
reply