The researchers used an AI system trained to identify comments in the dataset that were likely to violate Meta’s current […]