OpenAI and Google Harvest YouTube Data to Train AI, Reports New York Times

OpenAI has utilized more than a million hours of YouTube video transcriptions to train its latest artificial intelligence model, GPT-4, as reported by the New York Times. This extensive dataset was processed using OpenAI's Whisper audio transcription model, enabling the accurate conversion of spoken content into text. The initiative saw significant contributions from key OpenAI personnel, including President Greg Brockman, who was actively involved in the video collection process. Google also transcribed YouTube videos for text harvesting, highlighting a broader industry trend of leveraging publicly available data to enhance AI capabilities.