Personally identifiable information has been found in DataComp CommonPool, one of the largest open-source data sets used to train image generation models.

According to a job listing, Microsoft is launching a project to estimate the influence of specific training data on the media AI models create.