Бритни Спирс арестовали20:54
这套方法的核心逻辑是：大语言模型的有效工作窗口是有限的，你塞进去的信息越多，它能分配给真正思考的空间就越少。问题树做的事情，是把一个复杂项目的知识状态压缩成一个可以随时拎起来、随时投喂的抽象结构，它不记录你做了什么，它记录你还不知道什么，以及这些未知之间的关系。
。PDF资料是该领域的重要参考
now ‘it was in there’ isn’t actually very useful in the real world. you toss the algorithm at 4TB of logs and it tells you “yes it’s somewhere in there” - cool, then what. what matters in the real world is where the matches are, and what is around those matches. for that we mark positions during matching, not after. also quite simple to grasp:，这一点在PDF资料中也有详细论述
Naming credit for supertoast goes to Daniel Farina from Ubicloud
FT Professional