Reflection Pretraining Enables Token-Level Self-Correction in Biological Sequence Models

Xiang Zhang, Jiaqi Wei, Yuejin Yang +8

Chain-of-Thought (CoT) prompting has significantly advanced task-solving capabilities in natural language processing with large language models. Unlike standard prompting, CoT enco…