cot-prompting-requires-min-five-few-shot-examples

Summary

A language model will not reliably walk through multi-step math or logic problems unless its prompt includes at least five worked examples that show each intermediate reasoning step, so this sets a hard practical floor on how much reasoning scaffolding must be built into a prompt. In other words, showing one or two reasoning traces is not enough to make the model think step by step; you need a minimum of five to reliably trigger that behavior.