Also low priority, but there might be signal in metrics quantifying e.g., the amount of overlap between words used in the prompt and in the completion. Good to start from TextDescriptives only, I think, but keeping this open in case we decide to expand feature sets later.
Also low priority, but there might be signal in metrics quantifying e.g., the amount of overlap between words used in the prompt and in the completion. Good to start from TextDescriptives only, I think, but keeping this open in case we decide to expand feature sets later.