From Bloomberg: 2024-04-09 00:00:00

OpenAI’s CTO initially gave evasive answers when asked about using YouTube videos to train AI systems, showing discomfort with the topic. However, a New York Times report revealed that in reality, OpenAI did use over one million hours of YouTube videos to train its AI using a speech recognition tool called Whisper.

The company utilized conversational text from transcriptions to train GPT-4, their large language model powering ChatGPT. This approach showcases the extensive data harvesting and AI training methods employed by tech giants like OpenAI. The use of YouTube videos for training highlights the innovative techniques utilized in developing advanced AI systems.



Read more at Bloomberg: OpenAI Tapping YouTube? Big Tech’s Trapped in a Glass House of Its Own Making