We're very excited about the potential here, but we haven't landed on particular implementations yet. We are trying to harness the extraordinary creativity of generative AI models that allow you to create a poem or something—to respond to an email or whatever else it might be—and, in some cases, to summarize or distill content that's on the web, with the value of having a variety of content that people can go to directly.
You may have heard or read about the notion of AI hallucinations. In some cases, these generative AI models are predicting the next most likely word or words, but that doesn't necessarily mean they are grounded to underlying factual content, so a combination of those two tools turns out to be really quite important. One of the things we're looking to do is to understand how best to ground the models in a way that harnesses their powers of creativity and distillation but also draws on facts about the world.