Least-to-Most Prompting

Least-to-most prompting is a reasoning pattern in which the model first decomposes a complex problem into an ordered sequence of easier sub-problems, then solves each sub-problem in turn, feeding earlier answers into later ones. Introduced by Zhou et al. in "Least-to-Most Prompting Enables Complex Reasoning in Large Language Models" (2022), it extends chain-of-thought by making the decomposition explicit rather than leaving it implicit in a single reasoning trace. It is particularly effective on tasks with clear prerequisite structure — compositional generalization benchmarks, multi-step math word problems, and symbolic reasoning — where solving a hard instance requires first solving simpler versions. The trade-off is increased token usage and a dependency on the quality of the decomposition step; a bad decomposition cascades into bad answers.