OpenAI just gave away the entire game | Hacker News Reader

OpenAI just gave away the entire game

theatlantic.com

46 pointslolinder25 comments

This whole debacle has me extremely confused. I am not particularly enamored with OpenAI, especially the Sam Altman helmed version of it - I have all the usual "Actually ClosedAI," etc. complaints about it. I was hoping the governance structure was enough to force them to be a unique company last year. I think Altman is a snake oil salesman that has failed upwards.

But the Sky voice does not sound like it was trained off of ScarJo, or even off of someone who sounded a whole lot like her. There are some similarities, but, the voice sounds as close to, say, Rashida Jones as it does ScarJo. Yes, Altman posted a dumb tweet referencing Her which obviously looks to be in very poor taste, and yes they obviously wanted ScarJo herself, but does that mean they should not be allowed to have a voice that sounds even remotely similar to her?

It is standard and accepted practice for scriptwriters or directors to have a preferred casting in mind for a role, and when that person isn't able to be cast for whatever reason, for them to pick a person that fits similar criteria. What is different here?

The voices are different enough that I do not believe anyone as close to her as she claims would confuse the two. I pulled up a episode of Jimmy Fallon where she was on and compared the voice to one of the clearest audio videos of the Sky voice from the 4o demos, and they have different pitches, frequently different cadences to speech, etc. There are some similar mannerisms - there's a reason people see the resemblance - but if you isolate the audio of the two I don't think the majority of people would listen to them and go "Yep that's gotta be the same person," especially without all this context biasing their opinions.

They pulled the voice because it sounds similar enough, even if you don’t hear it. There are also several long form articles about it this morning to confirm further that it’s resonating as a major tech story because of those similarities. Probably most importantly the “voice of Her” is probably the #1 voice you shouldn’t try to emulate without permission (with Samuel L Jackson next). It is wild this has happened honestly, but I don’t think people are wrong about who it sounds like.

They pulled the voice because of all of the controversy. I would wager they turn it back on after this is all said and done.

I'll take that wager. How much? I say we make it large enough to be interesting.

The only reason for using a voice that sounds like ScarJo is to create an association between the product and the character in "Her". There is no other reason.

And whether you think it's fair or not, intentionally trying to tie your product to a celebrity without their permission is not permitted by the courts. See the Midler case. See the Watts case.

Making fun of Darth Vader on Saturday Night Live is fair use. Using Darth Vader to sell your cars is not.

The Midler case is someone that sounded far closer to Midler in an ad that was drawing multiple parallels to Midler.

If it is distinguishable to be a different person, courts have ruled that it's fine, see https://law.justia.com/cases/federal/appellate-courts/F2/435...

You don't think being a flirtatious, newly invented voice-only artificial intelligence is a parallel?

Hm.

Your question is answered in the case I linked. But no, that level of parallel was not enough in the Sinatra vs. Goodyear case.

The Sinatra vs. Goodyear is completely different. In that case, the company had licensed the rights from the rights owner (which happened to be someone other than Sinatra.) If OpenAI had written Spike Jones a check, it would be similar, but they did not.

You had said there were no parallels in this ScarJo situation and I'm merely pointing out that there are.

> In that case, the company had licensed the rights from the rights owner

You keep bringing up the Midler case in other comments, but Ford had licensed the rights for the songs from the rights owner as well. What's different here? If the Sinatra case isn't relevant to all of this because the song was licensed, I don't see how the Midler case stays relevant.

https://opencasebook.org/casebooks/3008-property-in-the-age-... > Young & Rubicam had a license from the copyright holder to use the song.

I did not say there were no parallels. I said there were multiple parallels in the Sinatra case: The usage of the song's melody, the 'boot' in the product name, lyrics tying the two together.

The only parallel we have here is Altman referring to a movie about a female-voiced AI assistant when releasing a female-voiced AI assistant. I think any reasonable person would agree there are far fewer direct parallels in the OpenAI/ScarJo kerfuffle than the Goodyear/Sinatra one.

> but Ford had licensed the rights for the songs

That's a fair point - thank you for bringing it to my attention.

I disagree with your count on parallels, however. By licensing the rights to the song, the song's melody and lyrics were not at issue, leaving only the boot.

Is the boot equivalent to the fact that both characters are revolutionary, flirtatious voice-only female artificial intelligences? I don't think it's anywhere close, but I accept that you'll likely disagree.

REGARDLESS, I'm having a seriously hard time understanding how the closing statement in Midler-v-Ford isn't directly relevant to this case:

"We need not and do not go so far as to hold that every imitation of a voice to advertise merchandise is actionable. We hold only that when a distinctive voice of a professional singer is widely known and is deliberately imitated in order to sell a product, the sellers have appropriated what is not theirs and have committed a tort in California."

It seems very clear that ScarJo's voice was deliberately imitated in order to sell a product.

The sentiment I hate a lot is that this is inevitable because generative AI is far too valuable for the world. Like no one, even the anti-AI writers like OP even seem to want to interrogate this point. Somehow it's only naysayers in comment threads (like here on HN) who question it.

Doomposting from people in the good old op-ed pages who are stereotypically uninformed about the things they're writing about only further feeds the frenzy. Of course, questioning where the current developments pass into hype hurts the frenzy, so it's strange how the anti-AI writers do not seem to adopt this framing at all.

> Like no one, even the anti-AI writers like OP even seem to want to interrogate this point.

People have tried that a lot in 2020 soon after GPT-3/Dall-E 2 was released. It's a losing game. The value is very clear and gigantic.

I have the feeling that when something important is going on, like OpenAI being unethical, or like blm back then that is what made me think about this, is that then magazines, blogs, internet vips and similar then pass day after day increasing the level of nonsense to the point that then the whole starting ideas become a joke, this is already exaggerated

I hope with the ethical question of ai companies we don’t get to the level of bs of “if you’re white then you can’t laugh to memes having black peoples”

blm? black lives matter? please explain

yeah, blm, they started as a reasonable movement, but ended up with giving too much space (imho) to people who started saying bigger and bigger bs (imho), over time in order to have some microphone, i.e. white people are all bad, white people can't listen to black music, etc. and it made imho the whole movement look like a joke

Don't forget one of the co-founders using org funds for a mansion and them likely acting as a disruption tool by foreign intelligence agencies.

You say "the co founders" like it was ever anything more than a hashtag. There are literally thousands of organizations using the name started by entirely random people, many of which are legitimate charities and many of which are grifts

The entire game is normalize generative while collecting conversations. Whichever players achieve ecosystem lock-in during the paid phase will have the network effects needed to sustain them during the commodization phase. They will know our capabilities, our dreams, our even our secret desires. And they will amass inconceivable fortunes monetizing us to their advertisers.

will? they already do. what do you think facebook, or gmail, or reddit, or the myriad 3rd party data aggregators has been collecting for years?

https://archive.ph/AL81B

The other movie to watch besides "her" is https://www.themoviedb.org/movie/152795-the-congress The Congress (2013) Based on Stanisław Lem's novel "The Futurological Congress." Robin Wright, who plays a fictionalized version of herself, facing dwindling career prospects, agrees to sell the rights to her digital likeness to a film studio, allowing them to use a computer-generated version of her in perpetuity.

But the big reveal here: actors don't even need to sell this. AI will just take it and claim well it's different enough. In fact, if you are a B or C celebrity you might have your agent get you a million dollars or so for these rights during this grey period. Go ahead and take the million because a few years from now the price will be zero.

This is not their "entire game". It's just the current state of cutting edge AI. Transformers and deep neural networks work this way, and require a lot of data, and in the end it's just creating a blended space that can be searched.

I think in general people have opted for having free services in exchange for their data. Social media and online ads received no pushback from people, and everyone was and is more than happy to give away their data in exchange for a service. So this ship has sailed, so to speak, and now there is no infrastructure to properly credit people's data contribution, and so OpenAI is just walking down the beaten path and applying it to current AI tech.

OpenAI is doing what most social media has already done, which is take people's data and provide a product for cheap. I would assume if they were forced to disclose and compensate every single piece of data for training GPT4... first it might take many years, and second it would make the product too expensive and therefore not worth the effort.

That quote about jobs is scary tbh

nursing homes here we come

Reply on news.ycombinator.com