There are hundreds of people with similar voices. If any voice actor can pull the same accent than Ms. Johansson, it should be fair game, as long it was the original training material? Voices cannot be copyrighted or be exclusive, although I am sure Hollywood will try to copyright them in some point.
He kind of ruined that argument when he tweeted “Her” alongside the video. Pretty clearly drawing a line between the voice and Johansson’s portrayal in the movie.
Incredible, really. It would have been so easy to just… not do that.
There could be a bubble in terms of stock valuation, but the tools are definitely going to stay.
This could be kinda like the dot com bubble -- the Internet went on to become BIG, but the companies just went bust... (and the ones that strive are probably not well known)
I understand there's way too much out there, but I think there is at least some clarity about the landscape at present.
ChatGPT is currently king of the mountain. That could change, but right now that's how it is.
Google's Gemini and Facebook's Llama 3 are clearly in a tier below. The 100s of tools you are seeing are various mixed and matched technologies that also belong in this tier.
Claude (massive context) and Mistral/Mixtral (decent with no censoring/guard rails) are interesting for special cases. And if you're determined and want to put in the effort, you can experiment or self-host and perhaps come up with some capabilities that do something special that suits a use case or something you want to optimize for (although not everyone has time for that).
So I wouldn't say it's just all this one big swirl of confusion and therefore a bubble and due to come crashing down. There's wheat, there's chaff, there's rhyme and reason.
This is completely false. Claude Opus is significantly better than GPT 4.
> Mistral/Mixtral (decent with no censoring/guard rails)
These models have been heavily censored, I'm not sure what you're talking about. Community efforts like Dolphin to fine-tune Mixtral have some success, but no, Karen is definitely still hard at work in France, ensuring that Mistral AI's models don't offend anyone's precious fee-fees.
I think you're missing the forest for the trees here. You're right that Claude Opus is better, which I hadn't known, but I think in your zeal to make that point you're completely forgetting what my comment was about.
It's nevertheless true that there is a coherent landscape of better and worse models, and Chat GPT really does have separation from the other models as I mentioned above. I even mentioned that ChatGPTs position would be subject to change. My understanding is that this most recent version of Claude has been out and about in the wild for perhaps 2 months.
I feel like with even a little bit of charitable interpretation you could read my comment in a way that accounts for the emergence of such a thing as a new and improved model of Claude. So I appreciate your correction but it's hard to see how it amounts to anything more than a drive-by cheap shot that's unrelated to the point I'm making.
It's an exuberance bubble. Every tech company on earth is racing to "do something with AI" because all of their competitors are trying to "do something with AI" and they don't want to be left out of the excitement. The excitement and exuberance will inevitably cool, and then a new thing will emerge and they'll all race to "do something with that new thing."
Oh, the irony. Actors are afraid of being digitalized and used without their content, and the first B2C AI company digitalizes a soundalike of the voice of one of the first 3 AI movies, without her consent…
Her was a movie with an AI assistant who talked like a normal human rather than an intentionally clunky "bleep blorp" dialect that lots of other movies go with. They even make fun of this in the movie when he asks her to read an email using a classic voice prompt, and she responds pretending to be a classic AI assistant.
The new voice2voice from OpenAI allows for a conversational dialect, most prominently demonstrated in pop culture by the movie Her. Sam's tweet makes perfect sense in that context.
Sky's voice has been the default voice in voice2voice for almost a year now, and no one has made a connection to the Her voice until it started acting more conversational. It seems pretty obvious that OpenAI was looking for a more conversational assistant, likely inspired by the movie Her, and it would have been cool if the actress had helped make that happen, but she didn't, and here we are.
Also Juniper has always been the superior voice model. I just now realized that one of my custom GPTs kept having this annoying bug where the voice kept switching from Juniper to Sky, and that seems to be resolved now that Sky got removed.
> Sky's voice has been the default voice in voice2voice for almost a year now, and no one has made a connection to the Her voice until it started acting more conversational.
Let's take a parallel situation from around 20 years ago, and see how you feel about it. I'm going back that far as a reminder of what was long considered OK, before AI.
Except Britney Spears was not hired for the role. They hired a Britney Spears impersonator for the scene. They did everything that they could to make it look like Britney, and think it was Britney. But it really wasn't.
Do you think that Britney should have sued the Chucky franchise for that? If so, should Elvis Presly's estate also sue all of the Elvis Presly impersonators out there? Where do you draw the line? And if not, where do you draw the line between what happened in Chucky, and what happened here?
I really don't see a line between now having someone who sounded like the actress, and then tweeting the name of one of her movies, and what happened 20 years ago with Chucky killing someone who looked like Britney, then showing a license plate saying "BRITNEY1", and THEN saying, "Whoops I did it again." (The title of her most famous song at the time.) If anything, the movie was more egregious.
> Seed of Chucky, the off-the-wall fifth installment of Don Mancini's Child's Play franchise, was forced to include a special disclaimer about pop superstar Britney Spears
> This scene was included in promotional spots for the film, most specifically Seed of Chucky's trailer, but the distributing company associated with the film, Focus Features, made the decision to significantly cut the scene down and add a disclaimer. The disclaimer that ran with the promotional spot, which was altered to only show a brief glimpse of Ariqat as Spears, stated: "Britney Spears does not appear in this film."
There is a distinction between the image of a celebrity and their voice. The image of a celebrity is usually pretty cut and dried, it’s them, or obviously intended to be them. If the use of their image isn’t meant to be satirical, it’s problematic. The Crispin Glover/Back to the Future 2 case is a good example of non-satirical use that was problematic. Zemeckis used previous footage of Glover, plus used facial molds of Glover to craft prosthetics for another actor.
Voices…are usually not so distinctive. However, certain voices are very distinct—Tom Waits, Miley Cyrus, James Earl Jones, Matt Berry. Those voices are pretty distinctively those people and simulating their voices it would be obvious who you are simulating. Other celebrity voices are much more generic. Scarlett fits into this with a pretty generic female voice with a faint NY/NJ accent.
Open AI screwed up by taking a generic voice and making it specific to the celebrity by reference and by actually pursuing the actor for the use of their voice.
I don't think this is an apples-to-apples comparison.
The movie producers didn't produce a simulation of Britney's voice and attempt to sell access to it.
However you feel about an probably-unapproved celebrity cameo in a movie, it's not the same thing as selling the ability to impersonate that celebrity's voice to anyone willing to pay, in perpetuity.
If you go to Vegas, you can go to a wedding officiated by someone who looks like, sounds like, and acts like Elvis Presly. This is available to anyone. You can get the same actor to do the same simulation for another purpose if you're willing to pay for it.
The biggest difference that I see is that technology has made the simulation cheaper and easier.
And these people are known as "Elvis Presley impersonators." They don't pretend to be some obscure person you've never heard of, for very obvious reasons.
The biggest difference here is obviously one of scale. I don't think ScarJo would be threatening to sue you, the individual, if you did a voice impression of her for a talent show or a friends wedding.
That makes it weird, but it doesn't (itself) mean they literally used her voice. It just means they were inspired by the movie. It's not illegal to be weird.
Legally they don’t need to have literally used her voice to have broken the law, never mind violating many people’s basic sense of what’s right and wrong.
They don't? Because if it's true that they used a sound-alike voice actress for the actual model, I don't see how any reasonable complaint about that could stand. You can't ban people from voice-acting who have similar voices to other celebrities. There needs to be something more to it.
It's such a huge problem that it's only brought up in the context of someone (probably) doing exactly what it's designed to prevent... By some miracle, this actually isn't used to outlaw satire or put Elvis impersonators out of work. It's used to prevent people from implying endorsement where none exists.
I think it’s less the voice and more about how they went about it. They were apparently in negotiations with her and they fell apart. Then they tried to resume negotiations with her two days before the new model launched.
If it was just an actor, it might be a case of inspiration gone awry. But this particular actor sued Disney in 2021 after making a lot of movies and a lot of money making movies for them.
Deliberately poking a fight with a litigation happy actor is weird. Most weird is really benign. But this is the kind of weird that forces out of court settlements. It’s reckless.
Edit - mistyped the date as 2001. Changed to 2021.
That's a fair statement if you take the "Her" post out-of-context and without the corroborating retort from ScarJo and his history. Which, of course, is not possible and also pretty boneheaded itself.
This isn't some college kid with an idea and too much passion.
Perpetual benefit of the doubt given for every implication as though it’s happening in a vacuum is how humanity keeps putting megalomaniacs and sociopaths into positions of power and influence.
If we're going to pillory Sam Altman, it's important to do it for the right reasons. That was not a good reason. I really should not need to defend this principle.
Had the film Her used someone else as the AI voice that sounded like Johansson would there be complaints about the film using a voice that sounded like Johansson? Does it matter if producers try to hire her first? Because only Johansson has that voice? Johansson does not visually show up the film Her and if not for the film credits could the voice in that film be used to use identity her from hundreds millions of other possible women? ( I had no idea who did the voice acting and would never had known if not for this news.) Now if the owners of the film Her were to request OpenAI licence a character from their film (like licencing say C3P0 character from Disney) maybe there would be a case but an actor claiming they own a natural human "voice" I think is a stretch when there are thousands of people with similar voices. And she is visually never in the film that made that AI voice famous so it could be anyone in that film with a similar voice.
I don't know about complaints but Ms. Johansson might be able to win a civil suit in that hypothetical situation. It would depend on the facts of the particular case, particularly any evidence that the defendants acted in bad faith. I think a lot of technologists don't understand how burden of proof works in civil trials, or that there is no presumption of "innocence".
This test is not blind but YOU tell me which you think is similar to the openAI sky voice? And what does that tell you about likely court result for Johansson? And having reached this conclusion yourself would you now think the other actress Rashida Jones is entitled to compensation based on this similarly test? Because there are no other women with similar voices?
> He kind of ruined that argument when he tweeted "Her"
Why? The grandparent is not saying it's coincidence. Why is it not okay to hire someone who has a voice similar to celebrity X who you intentionally want to immitate? I mean if you don't actually mislead people to believe that your immitation is actually X - which would be obviously problematic?
Alright then, the solution is simple. All he has to do is name the actress that OpenAI -did- hire for the voice work, right? That would put any doubt to rest.
(Can't for the life of me recall if she sounds anything like Johansson; just putting her forward to tease her relative here. (Who is in the wrong in his arguments above.))
This is one of those “accuse a diver of being a paedophile” moments. Who knew Sam is a creep with a Scarlet Johansson obsession cooking up a voice model just like her on compute daddy Satya paid for (but books as revenue, 2000 dotcom style).
In back to the future II, Crispin Glover didn’t sign up to be George McFly so they used facial prosthetics and impersonation to continue the George McFly character.
He sued Universal, and reportedly settled for $760,000.
While not defending OpenAI or Altman, the caveat here is that this was a voice actor using their natural voice, not an actor impersonating scarlett johansson.
Setting a precedent that if your natural voice sounds similar to a more famous actor precludes you from work would be a terrible precedent to set.
> Setting a precedent that if your natural voice sounds similar to a more famous actor precludes you from work would be a terrible precedent to set.
Yes, but literally no one anywhere is suggesting that the voice actress used would be banned from work because of any similarity between her voice and Johansson's; that’s an irrelevant strawman.
Some people are arguing that there is considerable reason to believe that the totality of the circumstances of OpenAI’s particular use of her voice would make OpenAI liable under existing right of personality precedent, which, again, does not create liability for mere similarity of voice.
>Yes, but literally no one anywhere is suggesting that the voice actress used would be banned from work because of any similarity between her voice and Johansson's; that’s an irrelevant strawman
It's not. The original comment in this chain was drawing parallel to a lawsuit in which someone intentionally took steps to impersonate an actor.
This situation is a voice actor using their "natural voice" as a source of work.
If a lawsuit barring OpenAI from using this voice actor is successful, due to similarities to a more famous actor, that puts this voice actor's future potential at risk for companies actively wanting to avoid potential for litigation.
Suggesting a calming female persona as a real time always present life assistant draws parallel to a movie about a calming female persona that is a real time always present life assistant is not a smoking gun of impropriety.
Pursuing a more famous name to attach to marketing is certainly worth paying a premium over a lesser known voice actor and again is not a smoking gun.
Sky voice has been around for a very long time in the OpenAI app dating back to early 2023. No one was drawing similarities or crying foul and decrying how it "sounds just like Scarlett" ..
> Sky voice has been around for a very long time in the OpenAI app dating back to early 2023. No one was drawing similarities or crying foul and decrying how it "sounds just like Scarlett" ..
While you're right I should have chosen my words more carefully, a random reddit post with 68 upvotes doesn't really dispute the substance of my comment.
OpenAI has been plastered across the news cycles for the last year, most of that time with Sky as the default voice. There was no discernable upheaval or ire in the public space suggesting the similarities of the voice in any meaningful public manner until this complaint was made.
The Reddit post had a link to a Washington Post article. And what you think the substance of your comment was is unclear.
Most people don't use ChatGPT. Many people who use ChatGPT don't use voice generation. OpenAI's September update didn't have a demo watched by millions unless I missed something. Altman hyped the May update with references to Her. Some people thought the recent voice generation changes made the Sky voice sound more like Johansson. Some people gave OpenAI the benefit of the doubt before Johansson revealed they asked her twice. And what do you believe it would prove otherwise?
>Washington Post article. And what you think the substance of your comment was is unclear.
You mean this?
"Each of the personas has a different tone and accent. “Sky” sounds somewhat similar to Scarlett Johansson, the actor who voiced the AI that Joaquin Phoenix’s character falls in love with in the movie “Her.” Deng, the OpenAI executive, said the voice personas were not meant to sound like any specific person."
As I stated prior, and thank you for making my point, despite being publicly available for near a year, there was minor mention of similarities with no general public sentiment.
>Altman hyped the May update with references to Her
If by "hype" you mean throwaway comments on social media that general population was unaware.
Drawing a parallel to a calming persona of an always on life assistant from pop culture in a few throwaway social media posts from personal accounts such as "Hope Everyone's Ready" isn't hyping it as Her any more than Anthropic is selling their offerings as a Star Trek communicator despite a few comments they've made on social media.
Ambiguous "some people" overstates any perceived concern and "most people don't use ChatGPT" understates how present they've been on the news.
Mobile app, which heavily emphasized voice and has "Sky" as it's default voice The ChatGPT mobile application had over 110+ million downloads across iOS and Android platforms before the May
announcement.
If we assume that Scarlett Johansson is telling the truth, why would they try to resume negotiations with her two days before they launched the model? If they found a good actor whose voice sounds like Scarlett Johansson, that’s a great argument. But if they found a good actor whose voice sounds like Scarlett Johansson because the real Scarlett Johansson said no, that gets more questionable.
When they did all that and still promoted the launch by directly referring to a Scarlett Johansson role, it got even more questionable.
I’m not pulling out my pitchforks but this is reckless.
Could they be trying to avert possible negative public perception even if they believe all they did was 100% legal? If you have ample funds and are willing to pay someone to make X easier for you does your offer to pay them imply that X is against the law? If your voice sounds like someone famous now you are prevented from getting any voice acting work? Because that famous person owns the rights to your voice? Tell me which law says this?
I don’t know why you’re asking me those last three questions. First, I’m not a lawyer. Second, I didn’t make any claims that could make those questions relevant.
Instead, I’ll repeat my earlier claim - this was reckless. If they were trying to avoid a strong negative perception, they failed. And they failed with an actor who sued Disney shortly after they paid her $20 million to make a movie.
You asked the good question about why they may have acted as they did and I attempted to answer it. In hindsight based on results it may look reckless but decisions need to be judged based on that is known at the time they are made and the public reaction was not a foregone outcome. The openAI sky voice has been available since last September why was there no outrage about it back then?
This test is not blind but YOU tell me which you think is similar to the openAI sky voice?
> And they failed with an actor who sued Disney shortly after they paid her $20 million to make a movie.
OpenAI did not fail. They suspended the sky voice and backed down not to further anger a segment of the public who views much of what OpenAI does in a negative light. Given the voice test above do you seriously think OpenAI would lose in court? Would that matter to the segment of population that is already outraged by AI? How are journalists and news companies affected by AI? How might their reporting be biased?
> While not defending OpenAI or Altman, the caveat here is that this was a voice actor using their natural voice, not an actor impersonating scarlett johansson.
Drawing a parallel to a calming persona of an always on life assistant from pop culture in a few throwaway social media posts from personal accounts such as "Hope Everyone's Ready" isn't "selling it as Her" any more than Anthropic is selling their offerings as a Star Trek communicator despite a few comments they've made on social media.
I think the issue is intent. It's fine if two voices happen to be similar. But it becomes a problem if you're explicitly trying to mimic someone's likeness in a way that is not obviously fair use (eg parody). If they reached out to Johansson first and then explicitly attempted to mimic her despite her refusal, it might be a problem. If the other voice was chosen first, and had nothing to do with sounding the same as Johansson, they should be fine.
No, it is. Waits v. Frito Lay was a successful lawsuit where Tom Waits sued Frito Lay for using an impression of his voice in a radio commercial. https://casetext.com/case/waits-v-frito-lay-inc
It's not that simple. Actors have a right to protect the use of their likeness in commercial projects like ads, and using a "soundalike" is not sufficient to say that isn't what you were trying to do. The relevant case law is Waits vs. Frito Lay. The fact that OpenAI approached her about using her voice twice and that Sam Altman tweeted about a movie she starred in makes her case much stronger than if they had just used a similar voice actor.
This is not the case.
“ A voice, or other distinctive uncopyrightable features, is deemed as part of someone's identity who is famous for that feature and is thus controllable against unauthorized use. Impersonation of a voice, or similarly distinctive feature, must be granted permission by the original artist for a public impersonation, even for copyrighted materials.”
> There are hundreds of people with similar voices.
Voice *actors* act. It is in the name. The voice they perform in is not their usual voice. A good voice actor can do dozens of different characters. If you hire a voice actor to impersonate someone else's voice, that is infringement. Bette Midler vs Ford, Tom Waits vs Frito Lay are the two big examples of court cases where a company hired voice actors to impersonate a celebrity for an ad, and lost big in court.
So when a cartoon show hires a sound alike replacement voice actor so that the switch is hard to tell the former actor has a case against the show? Perhaps instead the show has a case against the former voice actor using that same character voice elsewhere such as in radio advertising to impersonate cartoon characters that are not licenced?
So the voice of the AI in the film "Her" who do you think has more rights to it being reused elsewhere in association with AI? The voice actor? The film owners? Why then the current news?
No, voices can be exclusive. One good example is Bette Midler, who sued Ford in tort for misappropriation of voice and won on appeal to CA9. 849 F.2d 460.
This test is not blind but YOU tell me which you think is similar to the openAI sky voice? And what does that tell you about likely court result for Johansson? And having reached this conclusion yourself would you now think the other actress Rashida Jones is entitled to compensation based on this similarly test? Because there are no other women with similar voices? What might support from friends and family of Rashida Jones be an indication of?
