Session 5: Data in the Age of Generative AI

The podcast features James Zou and several colleagues discussing data's role in generative AI, highlighting both opportunities and challenges. They explore AI-created data, its societal impact, and the improvement of training data efficiency. The speakers present five research projects, including quantifying synthetic data, its use for public impact and AI training, data attribution, and understanding the sources of creativity and capabilities in diffusion and language models. They touch on copyright implications, the value of data, and future steps, such as deploying synthetic data techniques and designing data markets for generative AI. The discussion also addresses the legal, policy, and economic perspectives of data, model collapse, and the nuances of synthetic data. A Q&A session follows, covering topics like generating sensitive transaction data and attributing creativity in AI models.