If there's 10s of millions of people in china looking to cheat on their exams using llama models... i figure the "market" will make suitable models happen pretty quickly.

You realize of course that Exams are also highly tuned for one specific thing, like "20th Century English Literature", "Linear Algebra", "Criminlology: Overview of the Prison System", and "Axiomatic Meta-logic".

Eigenvalues are all but guaranteed to show up in the the 2nd exam, and all but guaranteed not to be mentioned in the other three.

PS That last one is highly recommended, but its a little dense.

It doesn't interfere with the "TrueDepth" Face ID stuff because it IS the TrueDepth Face ID stuff.

https://ancillary-proxy.atarimworker.io?url=https%3A%2F%2Fsupport.apple.com%2Fen-u...

You called it an "IR emitter"... but its an IR camera.

This camera is also used for "attention awareness features" (and for this, yes it is constantly taking pictures.)

Is it quite the situation the GP suggested... no, of course not.
But yeah, it is pretty constantly taking IR photos of you when those features are on.

It's likely not doing anything with them beyond the functionality apple claims... but it is taking them, and it could be doing something more with them.

I am if the AI generated slop can only exist because the LLM owner ripped off the human creators to train the LLM.

More like the same way Napster and Limewire worked, where we all just picked the songs we want and copied them for free digitally over the internet, rather than paying for a CD to be manufactured and shipped around. Napster was just the market deciding ... right?

Except it was massive copyright infringment. Sure the CD distribution model might have been outdated in the face of a more efficient digital model, but that doesn't mean just copying all the work you wanted for free was a good solution.

The LLMs right now are like Napster -- they took all that copyright work for free and are exploiting it for profit. That's plainly wrong.

We need to get to a spotify/apple music world, where the LLM systems respects copyrights. Artists decide if their content may be used as training material or not, and they get paid for it if they do.

I actually do agree with you here. But I think youtube is a pretty shitty platform, that promotes *mostly* shitty content. Because the engagement and incentive model of ad supported systems is pernicious. We can do better. But that's a separate conversation.

Separately LLMs training on their own AI slop causes problems and LLMs can easily produce content at a rate that can overwhelm human output ; so even without a creative apocalypse caused by people not wanting to create; there may be a practical one if ai slop drowns everything else under a flood of shit.

Humans are special.

Agreed. The contents of your mind are beyond the reach of copyright.
But the contents of your computer memory and hard drives are not.
Memorize it all you want, but copies on your computer are legally different.

Well, yes, people ARE forbidden from doing exactly that.

For example, if you want to recite lord of the rings to a paying audience? you need license for that. Translate it to Klingon and distribute copies?... you need a license for that too.

Sure you can show off to friends in a bar or around the kitchen table all you like privately. That's all fair use. But try to commercialize it and distribute it, broadcast it, or perform it? Yes, people also need a license for that.

AI companies by and large are for profit enterprises copying data that doesn't belong to them and exploiting it for commercial gain by training LLMs on it. LLMs aren't people. It takes input, it runs an algorithm, it generates output. For it to have a model of LotR in there, either the model is getting passed in the prompt the user gave it (it's not) or its getting in there via the training on the copy of LotR and everything else that was fed in (yep this one). So copies were made to train, and some sort of derivative copies are persisted inside the model data too... which then are commercially exploited.

And even if it did work exactly same way as a human brain worked it still wouldn't matter, because its still a computer, and memory and storage, and those things are all treated by copyright law differently from a human mind.

Copyright has always existed to carve out a monopoly for human creators; because we as a society have recognized that the humans at the top of the creativity chain are valuable and need to be preserved.

LLMs aren't going to replace artists, but letting people use them to generate content by training them on the work the artists creates the same problem in the end as the original printing press.

The printing press didn't displace authors. LLMs can't either. But if you let the people who own the LLMs just rip off all the authors, its the same problem you'd have if you let the people who owned the printing press do it.

Copyright isn't born of a desire to prevent the use of technology, whether it is the printing press or its an LLM. It's borne of a desire to ensure that human creators don't get completely ripped off by the people who own that technology.

In order to train an LLM, a copy is fed into the LLM (even if it isn't stored in the LLM, numerous copies are made for the purpose of trianing during training). The purpose of that copy is to enrich the owners of the LLM, which is a commercial purposel. Under what license was that copy made?

Those ephemeral copies are copies. There's lots of (stupid and awful) legal precedents around this stuff. Console hackers and software modders get slapped with this stuff all the time. e.g. lawsuits that succeed with arguments like "By modding the software you violated the EULA, and therefore, while the copy on the CDROM remains legal, the copy you made to your disk drive, and the copy you made in ram is not licensed, and therefor in violation of copyrights"

And then there is the imprint that these training works make on the LLMs... sure they aren't copies. But they're possibly derivative works, again created for commercial purposes... and again... where is the license for that? Just ask chatGPT how lord of the rings starts, and then ask it what happens next... and it'll take you through the whole thing. Not in Tolkien's words (although it can mimic that too if you want)... clearly the LLM has a complete and detailed model of the entire Lord of the Rings trilogy. Is that model not derivative work? It could clearly and easily create an obvious plagiarized work from what it IS storing, where the only prompting required is that it be asked to produce it. (And the only reason it won't is because its been told not to (aka guardrails).

Oh... and don't say 'but its the same thing as a human'. It doesn't matter. Human's are special. A human can only infringe by what we produce as "output". The contents of our minds can't infringe copyright simply by existing no matter how perfectly we've memorized Lord of the Rings. An LLM is just a computer with storage, and its storage/memory contents absolutely can be subject to copyright simply by existing.

They didn't make cars illegal, but they've regulated them from her to hell and back, precisely because moving something that big and heavy that fast among the general public causes all kinds of problems if it were not.

From seatbelts, and airbags, to requiring operators to have training and permits, and limiting where they can be used, and what direction they can go, and where they can turn, and limiting how fast they can be used, etc, etc, etc.

Is your car example meant to imply that are in favor of regulating AI just as comprehensively and aggressively as cars?

" I was there back in the day of analog long distance, and I can assure you that reliability was nowhere near that."

For your home landline?

Or for the dedicated line you were contracted directly with Bell with to provide point-to-point connectivity for critical communications infrastructure?

They weren't offering lots of 9's on your consumer land line, but that didn't mean they weren't offering it.

"If a person could look at a picture and say "hey! I recognize that place..." then a computer can do it faster and more reliably."

It's a common mistake to think that is the same thing. It's not.

Doing something automatically, massive at scale and storing that information changes the nature of that thing.

If I saw you at the beach, that's fine right? If I took a picture of my kids as you walked by in the background, no big deal right. But I followed you home constantly taking pictures, and storing in them in a file on you, I think you'd probably have a serious problem with it. Or maybe I follow your wife or daughter... is that ok? You were in public. I took pictures in public. What's the big deal?

Now if I buy 1,000,000 cameras and constantly take pictures of everybody, and then instead of following you home, i just tag you as you move into and out of each camera and build my profile on your and your households every movement that way instead.... Is that better?

You don't have to be the sharer of the photo to be in a photo. Do you accept that you have no privacy simply because you stepped out of your basement?

Some people do; especially singles living in condos or apartments; they get a comcast xfinity modile plan that gives them access to the xfinity public wifi being broadcast from the comcast cable subscribers modems in the neighboring suites, and just live on the free public wifi. It's not as fast or reliable as your own internet (but your phone has a data plan when you aren't in range of a free wifi), and you can only connect a couple devices to it at a time, so there are limits. But if all you've got is a smartphone and a laptop that you need to get online and there's decent wifi signal where you need it, then its workable. I've known people who do this.

The point here however is there is nothing stopping say LG or Samsung etc from partnering directly with comcast to let their TV adflow/telemetrics use the xfinity public wifi (or even cellular network) whenever it can't phone home another way. The price of building this capability in is dropping steadily. I saw a cellular punch-card system pre-covid for example, that just came with pre-paid cellular for the life of the unit. Its a time-clock, it logs punch in/ punch out events and time-sync, it could get by on a handful of megabytes a year. The writing is on the wall.

It's already well within the realm of economically viable to have a TV with pre-paid cellular to enable forwarding viewing metrics and serving image ads. And it might already be economically viable to download and rotate a view video ads per week / month or whatever too if you are buying bandwidth at volume directly from the provider or in some sort of data sharing partnership with them.

Sure that works in rural montana, but most people live in the city, so the TV can just connect to the neighbors modem instead of yours.

Or, wait for it, they partner with comcast and include a cellular modem right in the TV. Its getting cheap enough, and it doesn't to be all that good or all that fast.

They won't let you stream netflix over that connection of course, you can provide your own network for that.

But by selling/swapping your usage data that covers the bandwidth for a hundred MB a month of low priority data transfer which is plenty to upload some meta data and download new targeted ads.

Your solution? Put your TV in a metal box? Might work. Hard to watch TV though.

True, within the game that you purchased in the first place.

You have no redistribution rights to any of it. So "forcing the game open source" doesn't do much beyond let the people who already own it mess with the source code amongst themselves.

It's much like GLDoom and other ports of doom and quake that require you to purchase "Doom" or "Quake" to get your "license to the assets", to be able to use the upgraded "GLDoom" engine.

It is entirely *unlike* GLDoom though because where ID has consistently made some form of Doom legally available to purchase, if the original publisher pulls the game distribution or goes bankrupt WHERE does anyone buy the original game to get a license to the assets from anywhere to use with the modified source?

The most obvious blocker is pretty straightforward.

The developers of a game themselves may have licensed all kinds of things to build the game that they can't just give away, even if they wanted to.

From the game engine to the art assets to the music to the voice acting to the trademarks on the name itself. The developer doesn't necessarily own any of it. They may even contracted out parts of their own game code subject to terms that preclude open sourcing it.

I do think there is some balance that can and should be found for abandoned/end-of-life games to allow for preservation and continuation, but I can see where open sourcing it is not simple - even when the original developer and publisher has completely gone out of business, and no one even knows who owns the rights.

Whoops - sorry to self reply -- i ended up editng out a paragraph commenting on the structure of your posted ai response.

The first paragraph, opens with filler (defining what a trade imbalance is), before saying it doesn't know. The 2nd and 3rd are irrelevant. And the 4th is likely untrue.