Personally identifiable information has been found in DataComp CommonPool, one of the largest open-source data sets used to train image generation models. Millions of images of passports, credit cards ...
Unlike one-off tools, Column From Examples records transformation steps so results remain stable as new data is added.