OpenAI Executive Choked When Asked If Sora Trained on YouTube Data

OpenAI executive dodges question about Sora video generator training data, sparking comparisons to previous awkward interview responses. Viral-worthy gaffe.

During a recent talk at Bloomberg's Tech Summit in San Francisco, OpenAI's chief operating officer, Brad Lightcap, was put on the spot by Bloomberg's Shirin Ghaffary about the training data used for the company's new Sora video generator. Lightcap's response was a convoluted attempt to avoid directly answering the question.

In his response, Lightcap emphasized the importance of understanding where data comes from and hinted at a potential future "content ID system for AI" that would give creators control over their content being used for training. Despite his vague statements, Lightcap seemed to inadvertently confirm that Sora was indeed trained on YouTube data.

This incident mirrors a similar awkward moment involving OpenAI's chief technology officer, Mira Murati, who struggled to provide a clear answer when asked about Sora's training data in a previous interview with the Wall Street Journal. Murati initially stated that publicly available and licensed data were used, but when pressed about YouTube videos specifically, she admitted uncertainty.

The comparison between Lightcap and Murati's responses has sparked speculation that YouTube videos were indeed part of Sora's training data, despite the lack of a definitive confirmation. Financial journalist Rani Molla humorously pointed out that Lightcap's response essentially confirmed the use of YouTube data, leading to further scrutiny of OpenAI's transparency practices.

Overall, these incidents highlight the challenges and complexities surrounding AI training data and the need for greater transparency in the industry. As OpenAI continues to develop advanced AI technologies, the company will likely face continued scrutiny and pressure to be more forthcoming about their data sources and practices.