Our key finding is that these representations causally influence the LLM's outputs, including Claude's preferences and its rate of exhibiting misaligned behaviors such as reward hacking, blackmail, and sycophancy.
【启发】「情绪表征因果影响失控行为」这个发现，为 AI 对齐研究打开了一扇新门：与其设计更复杂的奖励函数或更严格的 RLHF，不如直接干预情绪向量本身。这启发了一种全新的对齐手段——「情绪工程」：通过调整特定情绪特征的激活强度，直接控制模型的行为倾向，而无需重新训练整个模型。这比 prompt engineering 更底层，比 fine-tuning 更精准。