It is interesting to me that Anthropic are more concerned about the "safety" of distillation training other LLMs, and not as much about an unscrupulously aggres

It is interesting to me that Anthropic are more concerned about the "safety" of distillation training other LLMs, and not as much about an unscrupulously aggressive goal-oriented solver that will do whatever it can to reach its goal, even if violates any kind of sandbox you might have reasonably expected.