Chain-of-Thought Prompting in Reasoning

What is the main benefit of using chain-of-thought prompting in complex tasks?

What is the advantage of combining chain-of-thought prompting with few-shot prompting?

What is the main idea behind zero-shot CoT?

What is the main limitation of hand-crafting effective and diverse examples in chain-of-thought prompting?

According to Zhang et al. (2022), what is the proposed approach to eliminate manual efforts in chain-of-thought prompting?

What is the advantage of using the prompt 'Let's think step by step'?

Why is diversity of demonstrations important in the proposed approach?

What is the main purpose of chain-of-thought prompting?

What is the main goal of Auto-CoT?

What kind of heuristics are used in Auto-CoT to encourage simple and accurate demonstrations?

Enables complex reasoning capabilities through intermediate reasoning steps
Can be combined with few-shot prompting to achieve better results on complex tasks that require reasoning

Adding all the odd numbers (9, 15, 1) gives 25 → The answer is False
Adding all the odd numbers (17, 19) gives 36 → The answer is True
Adding all the odd numbers (11, 13) gives 24 → The answer is True
Adding all the odd numbers (17, 9, 13) gives 39 → The answer is False
Adding all the odd numbers (15, 5, 13, 7, 1) gives 41 → The answer is False

Introduced by Kojima et al. (2022)
Involves adding "Let's think step by step" to the original prompt
Example: I went to the market and bought 10 apples... → Let's think step by step → Then you bought 5 more apples, so now you had 11 apples. Finally, you ate 1 apple, so you would remain with 10 apples.

Proposed by Zhang et al. (2022)
Eliminates manual efforts by leveraging LLMs with "Let's think step by step" prompt to generate reasoning chains for demonstrations one by one
Consists of two main stages:
- Sampling questions with diversity
- Generating reasoning chains to construct demonstrations
Encourages the model to use simple and accurate demonstrations by using simple heuristics such as:
- Length of questions (e.g., 60 tokens)
- Number of steps in rationale (e.g., 5 reasoning steps)