Ars TechnicaOpenAI holds back wide release of voice-cloning tech due to misuse concerns
Voice Engine can clone voices with 15 seconds of audio, but OpenAI is warning of potential misuse.
Paul Cantrell@arstechnica
I remember OpenAI holding back GPT from public release in a very similar fashion: it’s too powerful, concerns over misuse, etc. Then they went ahead and released it to the public anyway, with all the potential abuses barely mitigated and now materializing exactly as predicted.
At this point, I can’t view this sort of “It’s too powerful to be public!!” statement as anything but prerelease marketing hype.
To be clear, I’m very much in favor of ethical non-release of dangerous tech.
If you’re going to do that, the way to do it is not to send out a press release. It’s to keep your dangerous discovery somewhere between low-key and confidential, and get your research community working to develop mitigations and countermeasures •before• your creation wanders into the view of investors and militaries.
2xfo@inthehands
I keep thinking about the final episode of Silicon Valley and saying to myself "if only"
I keep thinking about the final episode of Silicon Valley and saying to myself "if only"
@RnDanger
Still haven’t seen the series! One of these days — although it keeps feeling like it’s going to hit so close to home it’s just annoying
Still haven’t seen the series! One of these days — although it keeps feeling like it’s going to hit so close to home it’s just annoying
@inthehands
I feel like Mike Judge does a great job showing the absurdity while it represents things very accurately.
He's an engineer, I've learned. He animated Beavis and Butthead to scale realistically with the mini Mart as they walked towards it, for example
I feel like Mike Judge does a great job showing the absurdity while it represents things very accurately.
He's an engineer, I've learned. He animated Beavis and Butthead to scale realistically with the mini Mart as they walked towards it, for example
Misuse Case@inthehands @BernieDoesIt OpenAI sent out a press release because remember their whole thing is doing a bit about how they’re worried about creating Skynet, when they know very well that’s not what they’re doing.
@MisuseCase
Yes, my point exactly
Yes, my point exactly
John Wilson@inthehands @arstechnica lol someone released a pretty damn good voice cloner today with source available
Saw it on hacker news
Worked with 2-3 seconds of reference audio, the output is immediately recognizable although a bit staticy.
Static doesn’t matter though, I bet it’s good enough for phishing
Karl Heinz Häsliprinz@inthehands @arstechnica They are weirdos. Voice cloning already worked pretty darn well with Open Source tools about a year ago (tortoise tts)... Didn't try the never models, but... Idk what OpenAI could do to be even more dangerous.
The only thing missing was a fool proof UI.
The only thing missing was a fool proof UI.
Howard Cohen@arstechnica potential misuse? Can someone explain to me how this tech can be anything other than misused? What is the ethical use for it? What use case exists for it that isn’t contrived?
Brent "the birdman" Loon@arstechnica I think people should need to fill out a request form to gain access. And only people who are making good jokes are allowed.
Tom Bellin 
@arstechnica I'm sure interested parties can gain access to this super dangerous technology by paying OpenAI a lot of money.
What a joke openAI has turned out to be. A cliche. Mad scientists building Frankenstein's monster over and over again. And then going "Whoops! Oh, well."
Sterling Hammer 
@arstechnica there’s no reason for this tech to exist.