Earlier quoted context omitted.
You go to court and see what happens.
Oh so its like patents
Actually that's easy: it wouldn't be profitable, yet.
111–120 of 461 posts
If LLMs are the ultimate remix machine, then is anyone with a RAG a digital DJ? One can't help but wonder what theft even means any more, when it comes to digital information. With the (lack of) legal precedent, it feels like the wild wild west of intellectual property and copyright law. Like, if even a superstar like Scarlett Johansson can only write a pained letter about OpenAI's hustle to mimic her "Her" persona,…
Copyright seems to always have one or another wild wild west going on. Maybe you are in the wrong place if the world constantly jumps and kicks from under you trying to throw you off?
Earlier quoted context omitted.
Seems silly. What if I train my model on somebody who sounds like a somebody?
I mean, it’s the same as trademark disputes; legal standards will slowly be cobbled together from statutes, regulations, and random judges setting precedent. “Confusion in the marketplace” seems like a potentially relevant term — accidentally producing a product similar to an existing person’s voice is one thing, but publishing it in a manner and/or context that makes it seem like that person recorded the lines is so…
Earlier quoted context omitted.
AI will be able to generate imperfection too.
That's convenient, right? AI will be more than perfect and also be able to be imperfect if you want it, it'll just do anything you want. Here's a better prediction: Almost no one wants to listen to whatever low effort trash gets pumped into an advanced TTS.
>I remember when OpenAI practically cloned Scarlett Johanssen's voice... There is absolutely zero evidence for this. I find it infuriating that this keeps being stated as a fact. So they go and hire a voice actor and clearly use her voice to train, but then they also scrape Scarlett Johansson from youtube and splice it into the training data to make the voice a bit more like hers? Really does that sound realistic?
Yeah it does. Motive: Altman had some weird boyish thing for her and they asked her first, she said no. Means: Lots of available data to use from her movies. They probably trained a model first without releasing it just because it’s ridiculously easy. Especially for OpenAI. Opportunity: AI is astonishingly good at laundering and remixing without exposing the training set, for previously-unseen levels of plausible den…
They just about manage to make a good multimodal transformer that can generate audio and you expect that right away they can also interpolate in latent space? How does that actually work? It's not so simple. What benefit do they have from training on Scarlett Johansson's data, because they sure as hell have a big risk. They clearly hired a voice actress and they clearly told her to sound like Scarlett Johansson in "Her", and the end result perfectly fits with that. The voice doesn't "uncannily sound like SJ", no it just vaguely resembles her voice and mostly just mimicks the mannerisms from the movie. For me this is a perfect example of Occam's razor. One explanation is simple and realistic. The other explanation requires significantly more advanced AI control than OpenAI has claim/demonstrated, and it requires Altman to be so obsessed with the SJ idea that he goes out of his way to secretly train on her voice, risking legal exposure, while still hiring a voice actress.
It's all fun and games until someone produces a recording of somebody else saying something incriminating and it will be used in court. This is the part of AI I hate.
It'll be bad for a few years, but surely at some point it'll become inadmissible in court because it's too easy to fake, right? But then what do we do, if video and audio footage is inadmissible?
Earlier quoted context omitted.
Stolen voices matter because what's being stolen here is the authors likeness, his reputation that he's build in the YouTube tech space and used for commercial products he had already reviewed. They chose exactly his voice for that reason. While AI voices will aesthetically be indistinguishable or even preferable they aren't going to carry any reputation or authenticity, which by definition is scarce and therefore va…
Again: this literally only matters currently in people trying to steal a voice. There's already VTubers who's whole visual identity is synthetic. Why wouldn't the same happen in any other space where performance can affect the perception of content, but you can now simply engineer the performance? Like I said: give it 5 years and you'll have influencers who no one has ever heard the voice of, because they don't make…
No. Their assets are synthetic. Their identity is not. Their value and identity is in the real relationship and trust the creator builds with the audience. And that is what the company in question here tried to imitate. They didn't copy Jeff for his angelic voice, they copy him for the trust implicit in his reputation, name and 750k followers. It's the same reason Chris Pratt still gets hired to do commercials for millions despite the fact that any acting student would do it for 500 bucks and why openAI tried to copy Scar Jo in particular.
All these performances can be done by anyone for nothing, but they have commercial value because of the reputation of the person that's lending their likeness to them. And that's of course also why this is blatantly illegal by the way, we have personality rights and you can't shill your product with someone's likeness without consent, whether that likeness is AI generated or reproduced in some other way.
It’s the Wild West and will be for some time but I agree, they should have the decency to use only voices of the dear departed. The library should be open source and hosted on GitHub. the talking dead seems like a good name for it. Obviously we will have to put it to a vote among the living.
That's even worse imo. There's a reason nearly every major world belief system includes a proscription against necromancy and this is exactly what they mean. The living should not speak with the mouths of the dead.
If LLMs are the ultimate remix machine, then is anyone with a RAG a digital DJ? One can't help but wonder what theft even means any more, when it comes to digital information. With the (lack of) legal precedent, it feels like the wild wild west of intellectual property and copyright law. Like, if even a superstar like Scarlett Johansson can only write a pained letter about OpenAI's hustle to mimic her "Her" persona,…
what theft even means any more They dragged the term through different phases, but that’s just projection of will. Theft is undefined for objects with .copy() interface. It’s still there when you look at it. People have to adjust expectations, not laws. Computers replaced computers, now voice acting replaces voice actors. Your popularity doesn’t mean anything really and wouldn’t it be unfair if only popular could spa…
In other words, that's just the normal lifecycle of words in a language with an active speaker community. In any stage of history, the meaning of words is just the speaker community's projection of will.
Best I can do now is acknowledge that what counts as "theft" is a complicated topic and can't be decided by a binary "is said object still there after alleged theft has occurred?". I've benefited from some digital theft, naturally, so I might be biased to uphold my own morality but the kind of theft contemporary AI tech has enabled is something else entirely. Somewhere there is where I draw the line.
Recently, I introduced a few friends to the works of digital artist wlop. The immediate reaction was "Is that AI?". I can't help but feel offended in behalf of wlop. It doesn't help that they have made LoRAs out of his work. It's not so much the "theft" of techniques/concepts/etc. that enrages me but rather, the theft of credibility that a human is capable of this output. I imagine Jeff Geerling (and, to a lesser extent, maybe ScarJo) is enraged along similar lines. In this AI summer, other people are fighting for their livelihoods, other people are fighting for their credibility. And, of course, there's an intersection of people whose credibility is their livelihood.
Note that in reframing it as theft of credibility, the owning party has been definitely injured to an extent. As in, said object (credibility) is no longer what it once was after alleged theft has occurred.
And I'm not trying to state some Universal Truths that I will debate to death. Again the whole point is that what counts as "theft" is a complicated topic. I'm sure if you spend a bit more brainpower, you can find analogies that will make me look like a hypocrite. I'm just seeing this community lately strongly signal towards preserving some "original" meaning of words in the belief that it will solve some problem or another and I'm tired of it; I have similar linguistic thoughts about the whole uproar on the term "hallucination" but that's for another comment thread essay.
> People have to adjust expectations, not laws.
I know this thread is about theft but this attitude is downright dangerous in general. People should expect laws to adjust, lest they become irrelevant. Quick example: it's not fair to tell workers to adjust their expectation in light of the emergence of the gig economy. Should they just expect their labor to be exploited then, moving forward? I say, absolutely not. Legislation should catch-up to uphold/strengthen labor laws. Replace "gig economy" with "AI" and we are sort-of back on topic.
Earlier quoted context omitted.
That's what happened with Scarlett Johanssen's voice in the OpenAI thing, right? Or at least I think it was a claim at the time.
She wrote a strongly worded tweet and people outraged for 0.03 ms and moved on to the next thing.