What I find somewhat heartening about this announcement is that the mainstream news coverage linked above actually goes into the societal impact and potential downsides immediately. It’s not any longer just “look what our tech wizards have cooked up for us now! Isn’t it cool?”
4 Likes
More likely we the viewers won’t notice such inconsistencies any longer because they’re just in pretty much every video we see. So we will adapt to the technology rather than vice versa.
Growing up in Germany everything on TV that wasn’t shot in German was dubbed. I knew that of course but the mismatch between mouth movements and speech never bothered me. I literally didn’t perceive it. Now, having spent decades only watching original language content, whenever I come across dubbed footage it is incredibly obvious to me.
I’m sure it is going to be the same once we are immersed in AI media. The little inconsistencies will just be so common and normal that we won’t care, and in some cases even lose the ability to see them as unusual.
7 Likes
they know this stuff is coming for their news jobs
3 Likes
Remember that each directly output example/attempt is made in one pass, like an unedited amateur video. Most of what most people actually watch is selected from among a wide range of scripts/pilots/etc., which undergo rewrites, and a long recording/editing/post-production process.
I still don’t understand why the so-called AI companies have been allowed to harvest vast amounts of copyright content to train their machine models.
And I still don’t understand what people think the AI is doing that isn’t covered by words like “reading,” “watching,” “listening,” “learning,” and “getting inspired by.” Show me the human artist whose neural nets were trained without drawing from a huge number of copyrighted works. No, the AI doesn’t work the same way. Yes, it uses a much larger volume of data (but much less computing power thinking about it). I’m sure this will get litigated and legislated many times in many places and I look forward to seeing how that plays out. Maybe we’ll finally be forced to acknowledge that the legal fiction that our official rules and informal concepts about copyright were ever sufficiently well defined needs revision.
The level of creativeness needed for a creative work is really low and I would expect that generating a prompt and selecting a particular result would probably qualify.
Certainly the caselaw so far hasn’t had any issues with granting copyright to computer generated works, the courts have refused to say that the computer programme is the “author”.
Depends on whether (and for how long into the development of future LLMs or other AIs) the courts think the AI is less of a person than this guy:
Granted I really like the idea of Microsoft and Google funding PETA to get monkeys declared legal persons to pave to way for AI litigation. 
YouTube.
Steal all the things and become so big everyone has to deal with you take your terms and pray you don’t alter them more. Don’t think that outright theft is some problem in the legal system. Record companies and publishers are actual, literal, every fucking day, thieves. In this world the way to be welcome in all the inner sanctums and feted in high society is to steal from artists and get rich rather than to create art.
Microsoft know what they are doing when it comes to making the law bend to their will.
3 Likes
Copying works. Type in “Italian plumber” etc and see the “AI” spew out its training data.
1 Like
Oh I strongly disagree. A book created last year by AI was denied a copyright. There’s a difference between computer generated (CGI) and AI (without getting into the issue of whether or not this is really AI—it isn’t). If the human labor consisted of generating a prompt and selecting a result, there is no way a copyright is getting granted for that. If you know case law to the contrary, I’d love to see it.
3 Likes
I’m very aware of that. I can do the same with a human artist. Whether or not each given instance is copyright infringement depends on multiple factors other than just whether they’ve seen Mario and Luigi in the past.
Also, remember that today’s LLMs are not actually storing their training data. They aren’t big enough for anything like that. They are learning features from it and then generating from that based on a prompt. It’s not spewing training data, because it doesn’t have that data stored. When I say “Italian plumber” what’s your first mental image? Because that’s what an LLM gives you, it’s first impression response.
1 Like
That is not actually true and has been proven false multiple times for multiple systems. It does spew out copyrighted training data. Regularly.
5 Likes
I don’t know that they’are allowed to, but they are just doing it, because tech companies (especially those like social media companies and companies working on things like AI) feel as if they can act with impunity, because they feel as if they are the most important, most valuable thing happening in the economy right now, and that they are measurably “improving” society. They feel as if taking “worthless” art found online their right, despite what copyright laws say, because they are doing something “useful” with it - turning into shareholder value (or increasing wealth for a few already wealthy investors, etc).
But, there IS a lawsuit happening right now…
https://catandgirl.com/4000-of-my-closest-friends/
No, I don’t see how it could. That just absurd… A person didn’t create the thing, a computer did. All the person did was enter some text as a prompt.
Because a computer isn’t a person…
The courts are correct. If you value a computer more than an animal, which has more in common with people, then, I really don’t know what to tell you…
That… straight up the enshittification process and the chokepoint process that Cory Doctorow talked about in his recent book with Giblin.
Can you? Because there is a direct copy that violates copyright, and then there is stuff covered under fair use as derivative works or parody or for purposes of criticism. AI is doing NONE of that. It’s just scrapping shit, and pooping out often exact replicas of other’s works… They are not people. They are not entitled to the same protections as people.
Are you saying that copyright protections should extend to our brains?
3 Likes
To provide some evidence for @robertmckenna’s response to you,
7 Likes
But… but… but… that’s just like… one! /s
2 Likes
Had he responded with that, my reply would have been a quite snarky “Tell me you didn’t read the article without telling me you didn’t read the article,” lol.
3 Likes
Honestly calling both the models and humans “neural nets” is reductive to the point of being inaccurate. They don’t have feedback loops like animals do, they have a separate training phase and application phase. The neurons have static numerical values as outputs, rather than firing at whatever frequency. They aren’t the same thing at all.
And as it happens, these neural nets don’t do any of the things you just said. The take in images and fit a complex hypersurface to them. They don’t attempt to understand any of the things in them – which is what reading, watching, listening, and learning all mean for humans, who don’t even see whole images without a lot of practice at it. And it certainly doesn’t get inspired by anything – it can’t extrapolate from its training data, only interpolate.
This comparison is frustrating. It’s a bad sign that people keep pretending AI is something completely different than what it actually is to excuse it lifting a huge amount of work without admitting its value. Let alone keep with the “too small to store training data” nonsense even after it’s been experimentally proven multiple times these giant models can reconstruct inputs.
8 Likes
It gets really philosophical real quick, doesn’t it? If I use a pencil, we don’t say well a pencil can’t own a copyright. If I describe a drawing to an artist, and they draw it, the copyright would be theirs without a contract between us. Lawyers are gonna have to argue through all sorts of things.
And then there’s all the copyrighted source material the models have consumed…
1 Like
Hm… I’m not so sure…
2 Likes
A.I. prompt “artists” are now complaining about their prompts being stolen.
What I wouldn’t give to have the power to grant people self awareness.
This whole thing is already eating it’s own tail here…
10 Likes
I mean, that makes sense, right? If you try to get as close as possible in multidimensional space to the coordinates suggested by a prompt, and you already have the data from an exact match occupying that space, you are going to reproduce that. It doesn’t matter whether the training data is stored as a jpeg or as a set of vectors, it’s there.
3 Likes
And it’s a problem either way
2 Likes
that’s what the company would like us ( and their competition ) to believe. it may be true, but it’s a company’s job to make money. telling the truth is often contrary to that goal
moreover, editing generated video to make it seem more seamless is just one of many ways to cheat their output.
unless they have invited reputable journalists in to play with it, and without restrictions report on their experience? but that doesn’t seem to be the case
well, i for one, only watch single origin video in it’s original uncut form. i pay good money, and i refuse to allow any so called editors to stand in the way of pure directorial vision. please zack snyder, set our world free!
3 Likes
