Peer-reviewedReviewed
Chain-of-Thought Prompting Elicits Reasoning
Showed that prompting a large model to emit intermediate reasoning steps before its answer unlocks multi-step reasoning that direct-answer prompting fails at, without any fine-tuning.
Executive summary
By putting a few exemplars that spell out step-by-step worked solutions into the prompt, the model imitates that format and reasons through arithmetic, commonsense, and symbolic problems one step at a time. The benefit appears mainly at large model scale and substantially raised accuracy on benchmarks like GSM8K math word problems. It made intermediate-computation prompting a standard, training-free way to get harder reasoning out of existing models.
Antecedents
Challenged, corrected, or evaluated by
- enables Language Models are Few-Shot Learners — Chain-of-thought emerges only at scaleDirect evidence
What it enabled — descendants
- extends Self-Consistency Improves Chain-of-Thought Reasoning — Self-consistency votes over many CoT samplesDirect evidence
- extends Structured Reasoning — ToT/L2M/PoT structure reasoning beyond a linear chainDirect evidence
- depends on DeepSeek-R1 — RLVR learns long chains of thoughtStrongly supported
- challenges Reasoning Correctives — CoT can be unfaithful; reasoning correctivesDirect evidence
- combines ReAct — ReAct combines reasoning with actingDirect evidence
- extends Solving Quantitative Reasoning Problems with Language Models — Minerva extends chain-of-thought to quantitative and mathematical reasoningStrongly supported