Top Tech Firms Used YouTube Subtitles for AI Training: Report

AI models were trained using YouTube video subtitles from Dhruv Rathee, Marques Brownlee, and PewDiePie, according to a tool supplied by the news outlet Proof News.

According to the site, top tech companies including Anthropic, Nvidia, Apple, and Salesforce trained their AI models using a dataset of YouTube video subtitles.

The publication claimed to have discovered subtitles for 173,536 YouTube videos taken from more than 48,000 channels, although it cautioned that the technology might provide misleading negative results.

Also Read: How to Summarize YouTube Videos with Google Gemini (Step-by-Step Guide)?

In addition to content creators like PewDiePie and Dhruv Rathee, some videos uploaded by tech critic Marques Brownlee were utilized to train AI. The videos were also featured in news articles and discussion programs from all over the world.

The fact that most of the recordings were from 2020 or before suggests a cut-off.

Companies that used video transcripts as a source for AI training material were criticized by Brownlee.

“Interesting fact: I pay a service (per minute) to have my videos more accurately transcribed before uploading them to YouTube’s back end. Thus, businesses that harvest transcripts are engaging in multifaceted theft of *paid* labour. Not great,” Brownlee wrote on Tuesday on X.

Also Read: Join YouTube Premium to test the AI-powered “Jump ahead” feature immediately

According to the site, Anthropic and Salesforce acknowledged utilizing training datasets that contained the video subtitles that were scraped, but they did not acknowledge any wrongdoing. Bloomberg, Apple, Nvidia, and Databricks did not address the claims with confirmation or denial.

When asked earlier in the year if ChatGPT’s creator used YouTube videos for AI training, OpenAI representative Mira Murati found it difficult to respond briefly.

Also Read: YouTube CEO Warns OpenAI Against Using Videos for Training AI Models