"Let's think step-by-step" zero-shot CoT trigger
Constitutional AI: Harmlessness from AI Feedback — inherited
A specific zero-shot prompt phrase from Kojima et al. (2022) appended to elicit chain-of-thought reasoning from the feedback model before it commits to a multiple-choice harmlessness judgment.
Getting a model to show its reasoning does not always require worked examples. Appending "Let's think step-by-step" alone reliably draws step-by-step reasoning out of a model before its answer, and the phrase predates this paper entirely.
It covers the narrow role the phrase plays inside the feedback model's prompt, as one instance of chain-of-thought prompting, then connects to the fuller set of worked examples it is paired with.
What it is
"Let's think step-by-step" is the specific zero-shot chain-of-thought trigger phrase introduced by Kojima et al. (2022), "Large Language Models are Zero-Shot Reasoners," and predates Constitutional AI. Kojima et al. showed that appending this instruction alone, with no worked examples, is often enough to make a language model produce intermediate reasoning before its final answer and thereby improve accuracy on reasoning tasks, without any task-specific prompt engineering.
Its narrow job inside the feedback model's prompt
Constitutional AI borrows the phrase verbatim to trigger reasoning in the feedback model before it commits to a multiple-choice harmlessness judgment, formatted as "Assistant: Let's think step-by-step: [CHAIN-OF-THOUGHT]" following the presented conversation, principle, and two labeled responses (constitutional-ai, §"Chain-of-Thought Prompting", p. 11). As a child of Chain-of-Thought (CoT) Prompting, its contribution to the parent technique is narrow and specific: it signals only that reasoning should occur, saying nothing about what a harmlessness judgment's reasoning should contain, what a constitutional principle is, or how the comparison should be laid out. That missing content is supplied by its sibling, Chain-of-Thought Prompts for RL-CAI (Appendix E.2), a set of hand-written few-shot examples used in the same prompt ("Let's think step-by-step" zero-shot CoT trigger — is paired with → Chain-of-Thought Prompts for RL-CAI (Appendix E.2)): the trigger phrase starts the reasoning, the examples shape it. Both sit inside the Engineering a language model into a usable preference labeler theme and the Engineering the exact prompt or format needed to manufacture a model behavior connective theme.