MIT researchers introduce weight-editing method to permanently excise dangerous knowledge from AI models · Digg