DeepMind's "​​Frontier Safety Framework" is weak and unambitious — AI Alignment Forum