This article is based on a single preprint paper with an evaluation sample size of only 100 prompts, and actual generalization capability still awaits validation at a larger scale.