Did Google use YouTube videos to train its new AI video model, Veo?

“We’ve utilized contented connected YouTube for galore years to marque YouTube amended for viewers and creators,” Voolich added.

Last week, YouTube announced that it volition beryllium integrating Veo with its Shorts diagnostic for users to make and upload six-second AI-generated clips. They volition besides beryllium capable to usage Veo connected YouTube to adhd AI-generated video backgrounds done the existing Dream Screen feature. YouTube is looking to rotation retired these AI features by adjacent year.

Tech companies person been criticised for not intelligibly stating what information has been utilized to bid their AI models. Earlier this year, OpenAI CTO Mira Murati drew flak for refusing to clarify if its text-to-video generator Sora had been trained connected YouTube content.

Concerns implicit AI-generated YouTube Shorts

Addressing concerns that AI-powered YouTube Shorts could beryllium perchance misused to make deepfakes, Voolich said, “We region harmful content, we besides leverage our proposal systems and monetisation tools to beforehand a steadfast ecosystem. With AI, we proceed to usage the aforesaid techniques.”

“We’ve besides updated our privateness ailment process to let individuals to petition the removal of altered oregon synthetic contented from YouTube erstwhile it realistically depicts them,” she further said.

YouTube has said that these AI-generated Shorts volition beryllium embedded with an invisible watermark based connected Google’s SynthID technology. They volition besides beryllium labelled arsenic having been generated utilizing AI tools. Though, users volition lone beryllium capable to spot these AI labels if they pat to grow the video description.

It is besides unclear if users volition beryllium capable to opt-out of watching AI-generated YouTube Shorts. While YouTube is exploring longer-duration AI-generated Shorts, Johanna Voolich explained that there’s a crushed for the existent six-second length.

“There is an elicit trade-off betwixt the prime of the videos we tin make and the clip limit. If we allowed a 20 2nd video, it would beryllium much analyzable to get it truly good. So, we deliberation for creators, grouping six-second videos 1 aft different volition let them to make a amended acquisition for users,” the apical enforcement said.

Other features coming to YouTube

YouTube volition soon rotation retired a caller diagnostic to fto users “hype” the videos posted by contented creators with a tap. The Hype enactment volition look beneath the existing Like fastener of a video. With much “hypes”, a creator’s video volition determination up a leaderboard of the apical 100 hyped videos. Users tin lone hype a YouTube video 3 times per week. The diagnostic volition lone look for videos posted by creators who person little than 5,00,000 subscribers.

When asked astir the 5,00,000 cut-off, Voolich told indianexpress.com, “We picked the subscriber threshold due to the fact that we wanted to person ‘Hype’ assistance smaller creators. So, the thought is to person their fans assistance them go successful. A batch of it is inspired from what we’ve seen with superfans successful APAC.”

On however the caller Hype fastener volition impact YouTube’s proposal algorithm, she said, “Hype volition not beryllium initially focused connected changing the main algorithm but putting creators connected a leaderboard to amusement who’s been hyped more.” Each state is expected to person its ain leaderboard of ‘hyped’ creators.

Though Voolich said determination was nary information connected however Indian YouTubers are adopting AI features, she stressed that “globally our volition is to little the obstruction for creators with these features.”