OpenAI didn’t copy Scarlett Johansson’s voice for ChatGPT, records show • Hacker News Board

Re: OpenAI didn’t copy Scarlett Johansson’s voice for ChatGPT, records show

#101

Post by hackerlight » Thu, May 23, 2024, 7:30 AM UTC

moogly wrote: post #68

Technically? Maybe not. In spirit? Sure as heck.

This could be confirmation bias.

Here's an alternative explanation that fits the same facts. Sam tweeted "her", not due to the voice, but due the existence of a conversational system that resembled the system in the movie. This primed people to hear Scarlett Johansson's voice in a generically cheerful female voice actor who was contracted over half a year ago. Scarlett's lawyers encourage her to write a public letter, helping Scarlett with the wording, in order to steer the public narrative, in order to place pressure on OpenAI to settle with her financially out of court.

Re: OpenAI didn’t copy Scarlett Johansson’s voice for ChatGPT, records show

#102

Post by numpad0 » Thu, May 23, 2024, 7:30 AM UTC

squigz wrote: post #15

pavlov wrote: post #13

Earlier quoted context omitted.

Surely there’s some kind of difference between “voice impression for a two-line cameo in one episode of an animated sitcom” and “reproducing your voice as the primary interface for a machine that could be used by billions of people and is worth hundreds of billions of dollars.” Is there a name for this AI fallacy? The one where programmers make an inductive leap like, for example, if a human can read one book to lear…

> for example, if a human can read one book to learn something, then it’s ok to scan millions of books into a computer system because it’s just another kind of learning. Since this comes up all the time , I ask: What exactly is the number of books a human can ingest before it becomes illegal?

Depends on similarities between existing data and generative outputs, so minimum is zero. Humans are caught plagiarizing all the time.

Re: OpenAI didn’t copy Scarlett Johansson’s voice for ChatGPT, records show

#103

Post by pests » Thu, May 23, 2024, 7:30 AM UTC

boxed wrote: post #78

SeanAnderson wrote: post #62

Earlier quoted context omitted.

Why do you feel OpenAI doesn't have a chance to win in court given the information presented in this WaPo article? It seems fairly conclusive to me.

Copying by hiring a sound-alike/impersonator is an implementation detail. They knew full well that they were trying to copy her voice. Then they asked. And they got a no. And then they did it anyway. I don't see how the implementation details matter at all.

It's their natural voice.

Are they not allowed to hire a person that sounds like... any famous person?

Aren't others in this thread claiming how similar voices can be and theyre not unqiue?

Re: OpenAI didn’t copy Scarlett Johansson’s voice for ChatGPT, records show

#104

Post by huppeldepup » Thu, May 23, 2024, 7:30 AM UTC

squigz wrote: post #88

joekinley wrote: post #29

I can't help but think that this was all planned. It is a very intricately planned, and geniously executed marketing ploy to make sure everyone knows about the company, the new release, that there is voice now, and even makes everyone look into it just to "see for themselves". Whether this was in with ScarJo in the loop or not, does not really change the outcome, but would be a nice information we probably will never…

I think you're giving far too much credit to those involved. Not everything is a plot.

It _does_ make me wonder whether spin doctoring and plotting will be available as a model in the near future.

Re: OpenAI didn’t copy Scarlett Johansson’s voice for ChatGPT, records show

#105

Post by Strawberry76 » Thu, May 23, 2024, 7:32 AM UTC

I think it's pretty obvious that OpenAI had decided at an early stage that the new voice should resemble the voice of SJ in "Her", regardless of which voice actresses they then contacted and in which sequence.

Re: OpenAI didn’t copy Scarlett Johansson’s voice for ChatGPT, records show

#106

Post by gnicholas » Thu, May 23, 2024, 7:34 AM UTC

ben_w wrote: post #95

ml-anon wrote: post #85

Then why pull the voice in the first place? If it goes to court I’m sure discovery will unearth a bunch of emails and slack messages pertaining to this as well as documentation about the make up of their training sets and casting and performance notes for the voice talent. Hopefully they’re under legal hold now.

> Then why pull the voice in the first place? Several possible reasons. 1. They're not actively malicious and have the emotional capacity to feel embarrassed 2. They know they were talking about her voice and making comparisons and that this is A Bad Look and they've had a terrible week PR-wise already 3. Might be a trademark violation, I'm not sure how that law works (in general let alone voice) but there's a reason…

1 is possible, 2 strikes me as unlikely (how bad of a week was it? I didn't get that sense). 3 doesn't make sense — this has nothing to do with trademark law. A logo is a 'mark' that identifies a brand, but a voice is not a 'mark.'

Re: OpenAI didn’t copy Scarlett Johansson’s voice for ChatGPT, records show

#107

Post by pfannkuchen » Thu, May 23, 2024, 7:36 AM UTC

Ariel_Asura wrote: post #5

It seems increasingly difficult for common people to protect their voices, especially when even Scarlett Johansson can't manage it. As a part-time voice actor with a unique voice, I'm concerned about what I should do if my voice is used without permission and the company denies it. How can I protect myself in such a situation?

I don’t think this will be a concern for long. Either the tech isn’t good enough and it lacks emotive nuance to the point where human is still preferred, or it is good enough and there is no point in basing off a human actor in the first place vs using an original wholly fabricated voice or appearance.

If the tech actually works well enough to stand in for humans, I think we will very quickly see recording real humans in fictional pieces as old fashioned.

Re: OpenAI didn’t copy Scarlett Johansson’s voice for ChatGPT, records show

#108

Post by gnicholas » Thu, May 23, 2024, 7:37 AM UTC

abrookewood wrote: post #69

If they are so confident that they didn't copy her voice, then why did they pull that character/version?

I wondered about this also. Maybe they pulled it temporarily so that their side of the story could get out in circulation. Then they'll reinstate the voice after a while on the grounds that "as we all know, this was developed separately and in advance of any conversations with Scarlett".

If they had kept the voice then it might seem like they were reacting and possibly caught off-guard. Instead, it looks like they're stepping back, assessing the facts, and proceeding with due caution.

Re: OpenAI didn’t copy Scarlett Johansson’s voice for ChatGPT, records show

#109

Post by dingclancy » Thu, May 23, 2024, 7:38 AM UTC

serial_dev wrote: post #34

skilled wrote: post #18

The thing that worried me initially was that: - the original report by Scarlett said she was approached months ago, and then two days prior to launch of GPT-4o she was approached again Because of the above, my immediate assumption was that OpenAI definitely did her dirty. But this report from WaPo debunks at least some of it, because the records they have seen show that the voice actor was contacted months in advance…

I don't know, to me, it's just sounds like they know how to cover all their bases. To me, it sounds like they had the idea to make their AI sound like "her". For the initial version, they had a voice actor that sounds like the movie, as a proof of concept. They still liked it, so it was time to contact the real star. In the end, it's not just the voice, it would have been the brand, just imagine the buzz they would h…

Sure but Skye is still not SJ.

Re: OpenAI didn’t copy Scarlett Johansson’s voice for ChatGPT, records show

#110

Post by TeMPOraL » Thu, May 23, 2024, 7:40 AM UTC

numpad0 wrote: post #102

squigz wrote: post #15

Earlier quoted context omitted.

> for example, if a human can read one book to learn something, then it’s ok to scan millions of books into a computer system because it’s just another kind of learning. Since this comes up all the time , I ask: What exactly is the number of books a human can ingest before it becomes illegal?

Depends on similarities between existing data and generative outputs, so minimum is zero. Humans are caught plagiarizing all the time.

Plagiarism is not illegal, it's merely frowned upon, and only in specific fields. Everywhere else, it's called learning from masters and/or practicing your art.