Comment by docere

The issue with "adversarial" questions like the blood pressure one (which is open-sourced and published 1 year ago) is that they are eventually are ingested into model training data.