3 ms·Keeping AI Under Control Through Mechanistic Interpretability (MIT Talk)5 points by Lordarminius 3y ago