Dan Jurafsky

In this brief, Stanford scholars test a variety of ordinary text prompts to examine how major text-to-image AI models encode a wide range of dangerous biases about demographic groups.

Foundation models are often trained on large volumes of copyrighted material. In the United States, AI researchers have long relied on fair use doctrine to avoid copyright issues with training data. However, our U.S. case law analysis in this brief highlights that fair use is not guaranteed for foundation models and that the risk of copyright infringement is real, though the exact extent remains uncertain. We argue that the United States needs a two-pronged approach to addressing these copyright issues—a mix of legal and technical mitigations that will allow us to harness the positive impact of foundation models while reducing intellectual property harms to creators.

Assessing the accuracy of automatic speech recognition for psychotherapy