Fabien Roger comments on Memorizing weak examples can elicit strong behavior out of password-locked models

[Fabien Roger]

Thank you for catching that typo. It’s fixed now.