Apple, Nvidia, and Others Used YouTube Videos Without Permission to Train AIs

Companies like Apple, NVIDIA, Salesforce, and Anthropic, among others, used the following without permission: Data from over 170K videos on YouTube to train your generative AI models.This is an ongoing investigation Evidence Newsin partnership with Wiredpublished this Tuesday (16).

The report notes that tech giants are extracting data from “YouTube Subtitles,” a collection of transcriptions from 48,000 channels on Google’s video platform, without the owners’ permission. The material used did not include images from the recordings.

Popular YouTube creators like MrBeast, PewDiePie, Jackcepteye, and Marques Brownlee are among those whose content is used to train generative AIs. There are also education and learning channel materialsLike those at Harvard, MIT, and Khan Academy.

The list also includes the following videos: BBC, ABC News, NPR, Wall Street Journal This New York Times — the latter has already filed a lawsuit against ChatGPT’s developer, OpenAI. misuse of your texts. Even programs like this Jimmy Kimmel Live, The Late Show with Stephen Colbert This Last Week Tonight with John Oliver was used.

How do you know if a video is being used to train AI?

Those responsible for the survey are a Tool that lets you search if videos or materials from your favorite creators are being used to train generative AIsThe source is available on the Proof News website, just type the name of the channel in the search field.

According to the report, Salesforce and Anthropic confirmed they used the dataset but denied any wrongdoing in practiceNVIDIA declined to comment, while Apple did not respond to communications from the researchers.

It is worth noting that YouTube CEO Neal Mohan stated in recent interviews: Removing content from the platform to train AI violates the video service’s usage policiesThis also applies to transcripts collected by big tech companies.