New tool could help researchers probe how models “unlearn” sensitive training material

Read original article