undefined | Better HN

0 pointsnotatoad8mo ago0 comments

it seems like this is basically youtube's ContentID, but for your face. as long as you upload your "cameo" aka facial scan to them, they can recognize and control the generation of videos with it. if you don't give them your face, then they can't/won't.

"Consent-based likeness. Our goal is to place you in control of your likeness end-to-end with Sora. We have guardrails intended to ensure that your audio and image likeness are used with your consent, via cameos. Only you decide who can use your cameo, and you can revoke access at any time. We also take measures to block depictions of public figures (except those using the cameos feature, of course). Videos that include your cameo—including drafts created by other users—are always visible to you. This lets you easily review and delete (and, if needed, report) any videos featuring your cameo. We also apply extra safety guardrails to any video with a cameo, and you can even set preferences for how your cameo behaves—for example, requesting that it always wears a fedora."

0 comments

3 comments · 3 top-level

felixakiragreen8mo ago

Brilliant.

Until you have 2 people that are near identical. They don’t even have to be twins, there are plenty of examples where people can’t even tell other people apart. How is an AI going to do it?

You don’t own your likeness. It’s not intellectual property. It’s a constantly changing representation of a biological being. It can’t even be absolutely defined— it’s always subject to the way in which it was captured. Does a person own their likeness for all time? Or only their current likeness? What about more abstract representations of their likeness?

The can of worms OpenAI is opening by going down this path is wild. We’re not current able to solve such a complex issue. We can’t even distinguish robots from humans on the internet.

mvdtnz8mo ago

I'm an identical twin so immediately I can see a pretty stupid obvious problem with this.

neilv8mo ago

If this company's guardrails end up sufficiently working well in practice (note phrases like "intended", "take measures", and "preferences...requested", on things they can't do 100%)... there will be weak links elsewhere, letting similar computation be performed without sufficiently effective guardrails against abuse?

How do we prepare for this? Societal adjustment only (e.g., disbelieving defamatory video, accepting what pervs will do)? Establishing a common base of cultural expectations for conduct? Increasing deterrence for abusers?

1 more reply

j / k navigate · click thread line to collapse