差别

这里会显示出您选择的修订版和当前版本之间的差别。

到此差别页面的链接

en:large-language-model:plain-language-review:0-plain-language-review-learning [2026/07/10 16:17] – Create English version with Latin path (full-width to half-width conversion) ctbotsen:large-language-model:plain-language-review:0-plain-language-review-learning [2026/07/11 07:45] (当前版本) – ctbots
行 7: 行 7:
 }} }}
  
-[中文版](大模型:白话综述:0-大模型白话综述学习)+[中文版](cn:大模型:白话综述:0-大模型白话综述学习)
  
 I am a beginner in large language models, gradually learning and understanding them. But in the process, I've encountered quite a few confusions. Here I adopt the Feynman learning method, hoping to explain some basic concepts, development history, technical solutions, and engineering practices of LLMs in plain language, to consolidate knowledge and fill in gaps. The following is a comprehensive overview. I am a beginner in large language models, gradually learning and understanding them. But in the process, I've encountered quite a few confusions. Here I adopt the Feynman learning method, hoping to explain some basic concepts, development history, technical solutions, and engineering practices of LLMs in plain language, to consolidate knowledge and fill in gaps. The following is a comprehensive overview.
行 481: 行 481:
 In this application-layer decoupling process, the LLM doesn't need to play both judge and god of wealth in the same context. Each sub-Agent's Prompt is extremely clean, carrying only two or three tool skills. Tool call accuracy directly skyrockets from 40% for the single Agent to over 95%. In this application-layer decoupling process, the LLM doesn't need to play both judge and god of wealth in the same context. Each sub-Agent's Prompt is extremely clean, carrying only two or three tool skills. Tool call accuracy directly skyrockets from 40% for the single Agent to over 95%.
  
-[中文版](大模型:白话综述:0-大模型白话综述学习)+[中文版](cn:大模型:白话综述:0-大模型白话综述学习)