Repeating “Let Me Think” 200 Times Makes an LLM More Accurate
中文摘要
重复“让我思考”200次能通过无意义的标记解锁大模型的参数化知识，进而提升其准确性。
English Summary
Repeating "Let Me Think" 200 times enhances LLM accuracy by using meaningless tokens to unlock its parametric knowledge.
原文节选
Inside the strange science of parametric knowledge recall and why meaningless tokens unlock real memories Continue reading on Level Up Coding »