📄 中文摘要
在使用 AI 提示的过程中，许多人会面临一个被称为“提示漂移”的问题。用户通常会对提示进行反复调整，直到找到有效的版本并投入使用。然而，随着时间的推移，用户可能会忘记原始版本的具体内容和参数设置，导致无法有效回滚。为了应对这一挑战，作者决定构建一个类似 Git 的系统来管理和版本控制 AI 提示，从而避免混乱和不确定性，确保能够清晰追踪每个版本的变化及其效果。
我为 LLM 提示构建了 Git——因为 `summarize_v3_FINAL_actually_final.txt` 不管用
📄 English Summary
I Built Git for LLM Prompts — Because `summarize_v3_FINAL_actually_final.txt` Isn't Working
The issue of 'prompt drift' arises when users fine-tune their AI prompts until they find a successful version, but later struggle to remember the exact configurations used. This can lead to confusion when trying to roll back to a previous version after several months. To address this challenge, the author has developed a Git-like system specifically for managing and versioning AI prompts. This solution aims to eliminate the disorganization associated with prompt files and provide clarity in tracking changes and their impacts over time.