推荐cubefox · 2023-10
“The system I outlined above was inspired by this intriguing paper:”
推荐它,自己设想的系统就受它启发
语言模型自造编程题,经解释器验证后自我微调变强
论文探讨语言模型能否像围棋自博弈那样,自己生成有教益的编程题来提升表现。做法是让模型合成编程谜题及其解答,用 Python 解释器筛掉不正确的,再用这些经验证的合成题目微调模型,让它借助解释器“自我提升”。在公开可用的语言模型上,测试准确率提升到原来的两倍以上。
评论者说自己设计的“自生成问题加验证”系统受它启发,也有人把它与用单元测试代替证明的类似做法并列介绍。
“The system I outlined above was inspired by this intriguing paper:”
“A very similar approach (with some sort of unit tests instead of proofs) is described here in more detail:”