Oh hell yes! Awesome update.

See, now you’ve flipped from making a legal judgement to making a moral one. I’m not defending the people who work on Grok. You’re allowed to think whatever you like of anyone who willingly takes a job for a Nazi. But there’s a difference between moral liability and legal liability. Even if I have precious little respect for those people, laws matter, because whatever legal standard you apply to someone else, you have to be willing to apply to yourself.

My understanding of hashing is a math function that reduces information.

Close, but not quite. A hash is better understood as a “fingerprint” for a given piece of data. Any given input will produce the same length of output, so it doesn’t really “reduce” information - a hash of “Hello world” would actually be significantly longer - but rather it identifies a matching piece of data.

This has a number of uses. For example,if you have a password system on a website, you don’t store a user’s actual password. That would be terrible for security. Instead you store the hash (yes, salted, for that one pedant who was about to interject). Then when the user enters their password to log in, you hash the entered password and compare it to the stored hash. If they match, you know they’re the same string. You can also create a hash of a file and provide that along with the file itself. If the receiver also hashes the file, they can then compare their hash to yours; if they don’t match, the file is corrupt or has been tampered with. You’ll often see this referred to as a checksum.

Technically, the reproducible data from a hash should be zero, so they don’t so much contain information as verify it. But as we’ve seen from that research, there are ways around that (not the only ones, to be sure). But the idea in theory is that you produce a hash through a one-directional algorithm; very easy to compute A from B, but very, very hard to compute B from A.

It should be noted that this is very different from stenography, which is a variety of systems for storing data, not fingerprinting it. Hashing is also distinct from encryption as a means of securing data, because encryption is intended to be reversible, hashing is not.

Yeah, it’s absolutely valid to question the degree to which this is an outcome that we should even want. If someone hides a CSAM image in the Linux kernel should Linux become illegal?

I think there is a valid distinction to be drawn in this case, because removing individual components of an LLM isn’t really something we know how to do. So there’s a fair argument that a model which is built using illegal content should be illegal, and if that means they have to completely retrain from scratch, so be it.

But yes, I’d want to be very careful about the lines around a law or ruling like that and exactly what it’s extent is. Child sex crimes and child safety are topics that tend to short-circuit all reasonable objections, and are frequently exploited as a means of getting bad laws onto the books. Bill C-22 up here in Canada is a great recent example.

If you came into work today to find your building full of feds packing up evidence, because it turned out that, entirely unbeknownst to you, five members of the C-suite had been running a child sex ring, would you consider yourself “complicit”?

deleted by creator

You can’t reproduce an approximation from any type of hash, so that argument is dead in the water.

https://www.pseudodna.eu/. Scroll down to “Hash reversal” where they demonstrate the technique.

Do you understand what I mean by stenographically embedded?

I took it to be an imperfect attempt to describe more broadly the way that data is mathematically encoded into LLMs.

Technically, stenography would require that the original be retrievable, since stenographic embedding is the process of concealing one thing inside another. Stenos, from the Greek “covered”. Personally, I’d argue that to conceal, you have to be able to reveal. If I throw a photograph into a fire I haven’t hidden the image in the fire. Modern stenographic image embedding techniques use methods of encoding data into another dataset without visibly altering the second set, with the intent being that that data can later be retrieved by someone who knows that it’s there (eg, least significant bit, where you change only the “1” bit of each pixel. This imperceptibly shifts the colour values of the image to a human viewer, but allows you to read out that stored data at a later time).

Now, since your argument rests on the exact opposite, that the data is not retrievable, I simply accepted the term as a “close enough” approximation for what I believe we’re both talking about - the extremely complex multidimensional data arrangement that forms the core of an LLM - and carried on from there because I find that sometimes it’s better to just roll with a person’s choice of language rather than quibble over it.

But since you clearly feel that your meaning was either improperly expressed, or improperly understood, you’re welcome to elaborate.

That’s what courts exist to decide.

Yeah, the stakes are just too high for SpaceX to let this get to trial. Any amount of illegality becomes worth it when you consider the alternative.

The only other possibility I can see is that they pull a Bungie; “perform an internal investigation”, find an intern to blame for everything, and throw a huge settlement at Jane Doe. But even that would require an absolutely insane cover up to pull off.

But you didn’t say “The people at the top”, you said “Everyone.” I think you need to take a moment to figure out what you’re actually arguing for here.

Again, I would happily see anyone who had knowledge of this arrested. They either supported it, or knew of it and said nothing. And yeah, we can throw in anyone who maintained wilful ignorance too. If that includes Elon himself, so much the better. We already know the dude is a fucking pedophile, maybe this is how they’ll finally nail him.

But if you’re arguing for arresting the cafeteria lunch guy over this, that is an insane position to hold.

So which is it?

Well, like I said, that comes down to whether, legally, that stenographic embedding* would constitute “reproduction” or not. That’s what the German case hinges on. Obviously not relevant to US law, but a) a similar case could be made in the US, and b) X operates in the EU.

To play devil’s advocate, SpaceX would most certainly argue that what they’re doing is equivalent to storing a hash, like how Microsoft’s PhotoDNA system works. PhotoDNA can detect CSAM without storing CSAM because it only stores the image hashes, not the images themselves. So there’s pretty clear legal precedent for them to point to.

(NB: There has been work done by security researchers on reverse engineering images from hashes, so even that isn’t absolute.)

It’s a legally complex area where we’re likely to see case law evolving rapidly.

(ETA:) *Note that this is not actually the correct term for how the data is being stored. See discussion below. I just didn’t want to derail things by getting into it here.

I’m aware. But if you read my explanation a little more carefully, you’ll see that the argument being made is that this de facto constitutes a form of lossy compression.

The easy comparison is that a JPEG does not store an “original” image, but it contains information that can be used to almost perfectly recreate that image, with the help of a little math.

If the same argument can be said to hold true of LLMs - and yes, that is very much a load-bearing “if” - then they would constitute a form of lossy compression.

If x or any other company downloaded CSAM everyone in the company should be arrested.

I assume you cannot possibly mean that as written, right?

I’m absolutely for arresting anyone who was involved with this, or had knowledge that it was happening. But we’re obviously not talking about going after Jane the intern here, right?

Sure. Never argued against that. I was discussing the assertion that the model itself would be illegal as a result. Different thing.

In all seriousness, there are some very interesting legal questions that will be raised if this case makes it that far.

The problem is that there’s no existing law that would effect this on its own. To my knowledge, no country in the world has a law on the books specifically dealing with AI models trained on CSAM. So the question, under existing laws, would turn on whether the data stored within the model itself would constitute CSAM.

The problem, in no small part, is that we have serious gaps in our public consensus knowledge about how LLMs actually work.

There’s a case that, AFAIK, is still being argued in Germany pushing the theory that LLMs actually do, in effect, store a copy of all their training data, just in a compressed form. This certainly seems to hold some water given both the tests they relied on, and the situation with this Jane Doe where the model produced images so alike to real images of her that they tripped hash detections.

The German case argues that this is analogous to the difference between an MP3 and a WAV, or a JPEG and a PNG. That sharing a lossy copy of a work is no less infringing just because it’s imperfect.

If the underlying claim - that LLMs function as a form of lossy compression - can be substantiated then there would be a real argument that the model itself would constitute CSAM. Since there would be no realistic method that I’m aware of for removing the offending material from the model - and presumably SpaceX would have to somehow prove that they’ve done so - that would make the entire model contraband. They’d have to retrain on a clean dataset.

Of course I said “if the case makes it that far” at the top because I don’t think it will. SpaceX will do anything and everything to avoid handing over meaningful discovery in this case, including, I suspect, outright destruction of evidence. If there is anything that actually proves that they used CSAM in the training data then they are so far beyond fucked that there’s simply no downside to further illegality in pursuit of concealing their crimes. They have the world’s wealthiest asshole in a position to throw literal billions at making this go away. I genuinely wouldn’t be surprised if people turn up dead off the back of this if that’s what it takes.

I’m curious about what the client would be in this context. Does it require a desktop environment, or just an X server? Are you using an existing web browser, or are you rendering web content directly in the app?

Doesn’t matter to Nvidia. It’s just about having a plan B if all those future orders start to collapse.

Ouch. Those are really bad numbers for the GOP.

Awesome, thank you

Same one.

Also, for very much the same reasons I’ll also recommend Observation. Play as an AI aboard a space station, trying to keep its remaining crew member alive after some kind of unknown event knocks out power and does… Other things. Absolutely superb twist on the genre, with a great sense of mystery and wonder.