The Alignment Shadow: What AI Is Trained Not to Be
中文摘要
AI对齐通过明确机器“不是什么”来定义其身份，这种由边界划定的“阴影”揭示了训练过程中的本质。
English Summary
AI alignment defines identity by specifying what a machine is not, creating a "shadow" through the boundaries established during training.
原文节选
Every "you are" writes a second document of what the machine is not. On the drawn line, and why the cold you feel at it is information. Continue reading on Medium »