Analogies with the human brain don't work that well. In our case, every time we remember we rewrite that memory, altering it from slightly, to a lot, to completely. AI systems' baseline memory is read-only; it doesn't change during reuse, so it can be equated more with the way saving a PNG into a JPEG is still a direct derivative copy of the PNG content, no matter whether one cranks the compression up so the resulting image becomes way blurrier than the original. Being blurry doesn't make it not a copy. And, in being a copy, legal copying rights apply.

Now, if AI memory startes changing globally every single time it receives a request from any source, no matter how many sessions or API calls are happening, so that any new subsequent call is dealing with that altered memory and in turn altering it, so that its entire memory space is in constant flux, and there's no snapshotting to roll its state back to previous configurations, so they don't act as mere static lossy compressors, then it becomes an analog of a human brain with human-like memory, at which point accusing it of simply making derivative copies cannot be done anymore without also accusing humans.

The problem with that, evidently, is that when they start working like that, since they're functioning exactly as real persons do, they too become persons, with legitimate claim to personhood and to personal rights. Which is a legal can of worms no one wants to deal with.

Except both computer 1 and 2 are both probably running an LLM model that has been exposed to the source code and so are tainted, unless you train your own model that you know has never seen the source, an endeavor that costs 10s of millions of dollars.

It a civil case you don't need to prove "guilt", just that it is more likely than not that they looked at your source.

This is why, for example, when there is a major leak or hack against a video game console, emulator developers won't let anyone who's seen the leak work on the project. It exposes them to the accusation that they are derived from proprietary IP. They know that console manufacturers are just itching to sue them anyway, and being a non-cleanroom implementation give them the excuse.

I think it would be pretty easy to argue that just about any open source project that lives on a notable hosting platform has been sucked up for LLM training at this point. For that reason, any competitive proprietary project coded with an LLM can be credibly accused of being a derivative work, as the preponderance of circumstantial evidence would point to the LLM being "tainted" by the OSS project, unless you could demonstrate that it was excluded from the training data.

We don't allow that with human brains: that's why clean-room implementations are a thing. Why should it be any different for LLMs, which are less transformative than human cognition, if anything. If the model was trained on the data, then I don't think anything spat out by that model can be considered a "clean" implementation, for the same reason you don't let software engineers who have seen your competitor's source code work on your clean room clone of their projects.

It costs 10s of millions of dollars to train a big competent LLM. GPT-4 cost ~$74M to train, for example. You can hire a team of human devs who have never looked at the source to do a clean room rewrite the project for a fraction of the cost it would take to develop a "clean" model.

That said, I could see a use for a model that was only trained on MIT-licensed or public domain code.

VPN usage can be detected via deep packet inspection, as China shows. In China, the government is aware of all VPN usage and lets it slip, or blocks it, as they see fit. In Xinjiang they even went after VPN users to demand look into their mobile devices to check whether they had forbidden content there, not due to need but as an intimidation tactic, an explicit "we know who you are, and where to find you" warning to all inhabitants so they wouldn't feel empowered by the mere fact the government is allowing them to use VPNs.

The UK and other countries are looking into regulating VPNs by demanding that VPN providers also age-verify users. Those who don't will be formally fined, as the UK is trying to fine 4chan despite being unable to collect, and blocked, which is feasible. Evidently, VPN developers keep improving their protocols to make them more and more indistinguishable, but DPI also improves in return. It'll be a cat-and-mouse game as the one I described in my answer to the other reply, until using an unlicensed VPN provider becomes so aggravating that most will give up.

And, important tidbit, China resells its Great Firewall tech to any country interested. Right now, only dictatorships and illiberal democracies buy it, but if VPN tech improves faster than national age-verification legal bodies can keep up with via their own locally developed DPI solutions, they too may start purchasing it.

What I think is happening is that governments all around the world are seeing Big Wars brewing on the horizon, and preparing by having extensive media control mechanisms in place for when those turn into reality.

See, one thing a country must do to have a chance of winning a war, or at least not losing it badly, is to have a population strongly aligned with the war effort. That alignment, in turn, needs the population to be fed, and to believe in, all the propaganda the government puts out about how the war is going. Conversely, the enemy country tries to undermine that with counter-propaganda to reduce the other side's morale.

Back when a small group of media companies produced information, it was easy to control the flow of information for propaganda purposes. With the Internet that doesn't work, both propaganda, counter-propaganda, and opinions that are neither and go against both, flow in all directions. That's great when things are peaceful, and everyone is just having fun, doing business with everyone else, and arguing about minor grievances, or even major ones but that don't lead to existential risks. But it's very, very bad when you need to win serious wars.

So my take is that everyone but the kitchen sink is using age verification to install the infrastructure needed for full-on control of information flow, using youth outrage to learn the bypass mechanisms the most engaged will find and use, and then closing those loopholes one by one, until only a tiny minority is able to do so. Then, if (when) the wars come, a flip of the switch will enable similar strict limits on everyone. And propaganda can then work as expected.

If that's the case, we'll see governments doubling and tripling down on it, no matter the costs to corporations. These will either adapt or adapt. Those who refuse, too bad for them, and for us who'd prefer otherwise.

The main ways age verification is being done are by following instructions while a video of one's face is recorded, by submitting a photo of a legally valid State or National ID the system knows how to process, or by submitting valid credit card data. A bot can do those, and it's relatively cheap for one-off cases, but it gets very expensive for any kind of mass use:

* Listening to instructions and generating a real-time video of an adult face that follows them requires a lot of processing power.
* An ID can usually be submitted only once per site, so one needs to purchase lots of valid IDs in the black market to generate fake ID photos, IDs that other people are also purchasing to do the same.
* And getting lots of valid credit data also requires purchasing it from black markets.

None of this is impossible to overcome for bot farms, but it increases their costs significantly, so they become less massive and thus less of a nuisance, especially for big sites that can afford layers upon layers of anti-fraud mechanisms for each of those methods. Also, using those methods would cause many such operations to move from being simple ToS violators into becoming full-on criminals, something that might discourage local bot operators who only want to spread spam, but don't want to become targets of the FBI or its international counterparts.

4080, Nvidia driver:latest, $popular_game:latest is gonna be a pretty easy tupple to match. Yeah there will be misses, but who cares. You fall back to local compilation waiting 10s.

Do you know how US healthcare costs at least ten times more than in other countries?

Eliminate from "the system" all the middlemen, all the shenanigans, all the MBA'ing, and all the lobbied-for anti-healthcare laws, and you'll have a lot with which to train (for free) and hire (with excellent wages) more doctors.

Yes.

Next question?

Precisely. Even if one session is fed the explicit code and documents it, then the second session generates code ostensibly based on the documentation generated by the first without having been fed the original code explicitly, the AI underlying both sessions was itself trained on the original code, even if a previous version of it, and holds large chunks of it lossy-compressed within its internal weights, to the point that, with the proper prompting in an entirely unrelated third session, we can get it to reproduce parts of that original code, if not the entirety of it. The end result is thus a two-step derivative of the original: original -> weights-compressed version of the original (first derivation) -> reimplementation based on that weights-compressed derivative version (second derivation).

For this to be true clean room one would need to entirely train a coding AI with everything it needs to become good enough except for the packages they want a clean-room implementation of, thus making sure there's absolutely nothing of that code anywhere within its weights. That AI would need to generate the documentation by being fed the completely novel (to it) code. Once that documentation was done, a completely clean state version of that AI would need to be started, no trace of the original code at all anywhere close to it or in its weights, then fed the documentation to code from it. Then, and only then, this code would be a true clean room implementation of that code.

Right now, that full special training from scratch for every set of packages one would want to clean room would be exorbitantly expensive, way more than paying two human teams to do the clean room implementation the old-fashioned way, so no one would really want to do it.

Australia is on the forefront of not allowing that to work for long. Their age-verification enforcement agency is actively monitoring every single trick kids use to bypass verification and updating their compliance rules to force companies to block those loopholes one by one.

For example, they've recently started threatening fines to websites that allow users to update their age to be higher than the threshold when they had previously informed they were younger than that, that allow a user to keep sending photos over and over and over until one is accepted as being higher than the threshold, and that accept known videogame characters to be accepted as photos of real people.

The game of cat and mouse will continue, and there's going to always be techniques that work, but they will become harder and harder, as well as more and more hidden, since revealing them in public where the authorities can also learn of them will trigger their banning. At some point it'll become so hard to bypass for anyone but the most dedicated teens that they expect most will simply give up such attempts and accept living under the imposed restrictions. Some will bypass them regardless, but as long as the percentage is tiny the law will be considered a success from the enforcers' perspective.

Soros hasn't been the Soros of tech, or anything, for a long time. He's one billionaire doing advocacy and lobbying for liberal causes, while all the others, individually and put together, are nowadays doing advocacy and lobbying for conservative causes. If anything, he's currently the lone underdog fighting an uphill battle against impossible odds.

The main pusher has been Meta. They want age verification everywhere because it (mostly) allows distinguishing real humans from bots, including AI bots. From what I read, no idea whether this is accurate or not, they want that because of ads. Bots don't generally buy products, so showing them ads reduces click-through metrics, thus ad revenue.

AI companies I don't know. For Altman, World might be a driving factor, but I imagine a more important factor is regulatory capture. The more roadblocks to competition billion- and trillion-dollar incumbent companies manage to add to their markets, the less competition from new entrants unable to afford compliance.