⚡ LLMsID: cot

Chain-of-Thought

Q: Few-shot vs zero-shot CoT? Why does "Let's think step by step" work?

【Chain-of-Thought 核心考点 1】Few-shot vs zero-shot CoT? Why does "Let's think step by step" work?。技术解析：Definition: CoT (Chain-of-Thought) prompting: include intermediate reasoning steps in examples so the model decomposes complex problems into stepwise reasoning instead of jumping to the answer. The few-shot form shows reasoning chains in 2-8 examples; the zero-shot form appends "... 完整推导与工程落地见：https://talentme.airsota.com/resources/tech/roadmap/mindmap/LLM/card/cot

Q: How does CoT emergence relate to model size, and why do small models benefit little?

【Chain-of-Thought 核心考点 2】How does CoT emergence relate to model size, and why do small models benefit little?。技术解析：Definition: CoT (Chain-of-Thought) prompting: include intermediate reasoning steps in examples so the model decomposes complex problems into stepwise reasoning instead of jumping to the answer. The few-shot form shows reasoning chains in 2-8 examples; the zero-shot form appends "... 完整推导与工程落地见：https://talentme.airsota.com/resources/tech/roadmap/mindmap/LLM/card/cot

Q: How does self-consistency work, and why does majority voting improve accuracy?

【Chain-of-Thought 核心考点 3】How does self-consistency work, and why does majority voting improve accuracy?。技术解析：Definition: CoT (Chain-of-Thought) prompting: include intermediate reasoning steps in examples so the model decomposes complex problems into stepwise reasoning instead of jumping to the answer. The few-shot form shows reasoning chains in 2-8 examples; the zero-shot form appends "... 完整推导与工程落地见：https://talentme.airsota.com/resources/tech/roadmap/mindmap/LLM/card/cot

Q: Limitations of CoT? When is it ineffective or even harmful?

【Chain-of-Thought 核心考点 4】Limitations of CoT? When is it ineffective or even harmful?。技术解析：Definition: CoT (Chain-of-Thought) prompting: include intermediate reasoning steps in examples so the model decomposes complex problems into stepwise reasoning instead of jumping to the answer. The few-shot form shows reasoning chains in 2-8 examples; the zero-shot form appends "... 完整推导与工程落地见：https://talentme.airsota.com/resources/tech/roadmap/mindmap/LLM/card/cot

Q: How does CoT relate to test-time compute scaling?

【Chain-of-Thought 核心考点 5】How does CoT relate to test-time compute scaling?。技术解析：Definition: CoT (Chain-of-Thought) prompting: include intermediate reasoning steps in examples so the model decomposes complex problems into stepwise reasoning instead of jumping to the answer. The few-shot form shows reasoning chains in 2-8 examples; the zero-shot form appends "... 完整推导与工程落地见：https://talentme.airsota.com/resources/tech/roadmap/mindmap/LLM/card/cot

思维链 CoT

🎯Core Definition

CoT (Chain-of-Thought) prompting: include intermediate reasoning steps in examples so the model decomposes complex problems into stepwise reasoning instead of jumping to the answer. The few-shot form shows reasoning chains in 2-8 examples; the zero-shot form appends "Let's think step by step" to elicit stepwise reasoning. Key empirical fact: CoT ability emerges markedly only at ~

\geq 100B

parameters; small models gain little.

💡Use Cases

math, logic, and multi-step reasoning; combined with self-consistency (sample multiple CoT paths and majority-vote the answer) it boosts accuracy further; also the foundational idea behind o1-style reasoning models.

⚡Key Problems Solved

direct answer generation is inaccurate on hard tasks and errors can't be localized; CoT converts one big step into many small ones, giving the model explicit scratch space, reducing error propagation, and making reasoning inspectable and debuggable.

🎯5 High-Frequency Exam Points

Few-shot vs zero-shot CoT? Why does "Let's think step by step" work?

How does CoT emergence relate to model size, and why do small models benefit little?

How does self-consistency work, and why does majority voting improve accuracy?

Limitations of CoT? When is it ineffective or even harmful?

How does CoT relate to test-time compute scaling?

📖 In-depth Guide:📄 reasoning-and-cot →

Updated 2026-08-12

🎯

Test Your Knowledge: Practice Questions for "Chain-of-Thought"

Single choice pitfall questions with instant feedback and mistake tracking.

🚀 Start Card Practice ➔

← Previous CardDense vs MoE Next Card →Test-Time Compute Scaling

🔗 More LLMs Knowledge Cards

Agent & Tool Calling Alignment Tax & Preference Data Scaled Dot-Product Attention Attention Variants MHA/MQA/GQA