AI 正则表达式科学家:一个自我提升的正则表达式求解器
我构建了一个系统,其中两个大型语言模型(LLM)代理共同进化:一个负责生成正则表达式问题,另一个则学习如何解决这些问题。生成器分析求解者的失败,以创造出在其能力边缘的挑战。
结果是:自主发现了一套从简单模式到复杂正则表达式的课程,同时通过质量多样性档案确保了广泛的探索。
博客链接:https://pranoy-panda.github.io/2025/07/30/3rd.html
代码链接:https://github.com/pranoy-panda/open-ended-discovery
查看原文
I built a system where two LLM agents co-evolve: one invents regex problems, the other learns to solve them. The generator analyzes the solver's failures to create challenges at the edge of its abilities.<p>The result: autonomous discovery of a curriculum from simple patterns to complex regex, with a quality-diversity archive ensuring broad exploration.<p>Blog: https://pranoy-panda.github.io/2025/07/30/3rd.html<p>Code: https://github.com/pranoy-panda/open-ended-discovery