james.lucassen comments on Understanding the two-head strategy for teaching ML to answer questions honestly