Comment by conguy

Train a small model to preserve all decision-relevant information while becoming invariant to the human language used to frame that information?