Google Leverages YouTube Videos for Training Gemini and Veo 3 AI Models

Google is utilizing YouTube’s vast library of videos to train its cutting-edge AI models, including Gemini and the new Veo 3 video generator. YouTube’s extensive repository, accumulating over 500 hours of uploads each minute, provides invaluable real-world content. This move has garnered both admiration for its innovative approach and concerns over privacy and ethics, especially regarding the use of proprietary material. At the Cannes Lions festival, YouTube CEO Neal Mohan announced Veo 3’s upcoming integration into YouTube Shorts, enhancing user-generated content with AI-generated 8-second clips featuring synchronized audio. While Google confirms that YouTube videos are vital for improving AI capabilities—recognizing gestures, language, and pacing—this raises questions about user consent and copyright. Although Google promotes Veo 3 as a creative tool, critics highlight the potential for misinformation, pointing to instances of the tool creating misleading videos. Existing safeguards, like digital watermarking, are seen as inadequate against possible abuses in AI-generated content.

Source link