"The new controls are part of a broader batch of weekend updates meant to stabilize Sora and manage the chaos brewing in its feed. Sora is essentially " a TikTok for deepfakes," a place to make 10-second videos of pretty much anything, including AI-generated versions of yourself or others (voice included). OpenAI calls these virtual appearances "cameos." Critics call them a looming misinformation disaster."
"Bill Peebles, who heads the Sora team at OpenAI, said users can now restrict how AI-generated versions of themselves can be used in the app. For example, you could prevent your AI self from appearing in videos involving politics, stop it from saying certain words, or - if you hate mustard - stop it from showing up anywhere near the hellish condiment."
"The safeguards are welcome, but history of AI-powered bots like ChatGPT and Claude offering up tips on explosives, cybercrime, or bioweapons suggests someone, somewhere will probably figure out a way around them. People already have skirted one of Sora's other safety features, a feeble watermark. Peebles said the company is also "working on" improving that. Peebles said Sora will continue "to hillclimb on making restrictions even more robust," and "will add new ways for you to stay in control" in the future."
Sora introduces user controls allowing people to restrict where and how AI-generated versions of themselves appear, including bans on political content, forbidden words, and appearance preferences. The app enables creation of 10-second deepfake videos with voice, called "cameos," which has prompted concerns about misinformation and abuse. OpenAI acknowledges shortcomings such as a weak watermark and says it is working on improvements. Team leads promise ongoing enhancements to restrictions and additional control options. Critics warn that past AI behavior suggests determined actors may find ways to circumvent safeguards, keeping risks significant.
Read at The Verge
Unable to calculate read time
Collection
[
|
...
]