New Delhi: OpenAI, led by Sam Altman, has transcribed more than a million hours of YouTube videos to exercise its GPT-4 style of AI, according to a report.
The New York Times reported that OpenAI knew it wasn’t legal but “thought it was a fair use. “
“OpenAI’s president, Greg Brockman, took personal care to collect the videos used,” according to the report.
An OpenAI spokesperson told The Verge that the company uses “many sources, adding publicly available knowledge and partnerships for private knowledge” to its global competitiveness studies.
Google, which owns YouTube, had “seen unconfirmed reports” about OpenAI’s activity.
“Our robots. txt files and terms of service prohibit the unauthorized removal or downloading of content from YouTube,” the tech giant argued.
Last year, The Information first reported that OpenAI, now subsidized by Microsoft, had trained its AI models on Google-owned YouTube, deleting its data.
OpenAI “secretly used knowledge from (YouTube) to exercise some of its synthetic intelligence models. “
YouTube is the largest and richest transcription of images, audio, and text on the web.