| ID | Prompt Name | Text |
|---|---|---|
| None | Direct | Direct prompting. No specific prompt is used. Just the question and answer choices are the input to the model. |
| kojima-01 | (cot_trigger) Kojima | Answer: Let’s think step by step. |
| zhou-01 | (cot_trigger) Zhou | Answer: Let’s work this out in a step by step way to be sure we have the right answer. |
| zhou-01-ins | Zhou-instruction | Let’s work this out in a step by step way to be sure we have the right answer. |
| qa-10 | (instruction) Plan | First think step by step – describe your plan for how to get to the right answer, written out in great detail. Then answer the question. |
| qa-12 | (instruction) Articulate | Carefully read the question & work this out in a step by step way to be sure you have the right answer. Be certain to spell out your thoughts & reasoning so anyone can verify them. Spell out everything in painstaking detail & don’t skip any steps! |
| qa-13 | (instruction) Rephrase | Instruction: First let’s rephrase the question to be sure we understood it correctly. Second, let’s work this out step by step by spelling out our thoughts & reasoning so anyone can verify them. Third, make sure we have the right answer. |
| qa-16 | (instruction) Elaborate | Answer the following question through careful, concise step-by-step reasoning. First, complement the question with helpful knowledge and important additional facts. Second, generate sub-questions that are required to answer the original question, answer them until you can answer the original question. |
| qa-17 | (instruction) Converse | Create a dialog between a professor and a student. The student asks sub-questions to the question. The professor works them out in a step by step way and makes sure that the student understood how they got to the right answer. |
| refl-01 | (instruction) Self-critique | Answer the question, then critique the answer. Based on the critique, reconsider the other answer options and give a single final answer. |
Table 5: Used prompts and their corresponding ID and text.
Introduction: The art of asking questions is an essential skill that can lead to deeper understanding and effective communication. As AI-powered language models continue to evolve, their ability to understand and respond to various types of questions becomes increasingly important. In this blog post, we will examine a fascinating chart showcasing a collection of carefully crafted prompts, each designed to encourage a unique and effective approach to answering questions. By understanding the nuances of these prompts, we can unlock the potential of AI-powered question-answering systems to deliver more insightful, accurate, and helpful responses.
The Chart: The chart, titled “Table 5: Used prompts and their corresponding ID and text,” presents an array of diverse prompts, each with its own distinct objective. These prompts are divided into three columns: ID, Prompt Name, and Text. Let’s take a closer look at each prompt and its intended purpose:
- Direct: This prompt is a straightforward approach that provides no specific guidance to the model, relying solely on the input question and answer choices.
- (cot_trigger) Kojima and (cot_trigger) Zhou: Both of these prompts encourage a step-by-step thought process to arrive at the correct answer. The primary difference between them is the phrasing, with Kojima’s prompt using “Let’s think step by step,” while Zhou’s prompt uses “Let’s work this out in a step-by-step way.”
- Zhou-instruction: Similar to the (cot_trigger) Zhou prompt, this instruction emphasizes a step-by-step method to ensure the right answer is reached.
- (instruction) Plan: This prompt requires users to create a detailed plan outlining the steps they will take to reach the correct answer before actually answering the question.
- (instruction) Articulate: This prompt emphasizes clarity and thoroughness by asking users to carefully read the question and provide a step-by-step explanation of their reasoning, ensuring anyone can verify their thought process.
- (instruction) Rephrase: This multi-step instruction involves rephrasing the question to confirm understanding, working through the problem step by step, and finally making sure the correct answer has been found.
- (instruction) Elaborate: This prompt encourages users to complement the question with additional knowledge and generate sub-questions that need to be answered before arriving at the final solution.
- (instruction) Converse: This creative prompt involves a simulated dialogue between a professor and a student, allowing the student to ask sub-questions while the professor walks them through the problem-solving process.
- (instruction) Self-critique: This prompt asks users to answer the question, critique their response, and reconsider other options before providing a final answer.
Conclusion: As we have seen, these prompts offer a wide range of techniques to guide language models in their quest to deliver accurate and insightful answers. By understanding the unique characteristics of each prompt, we can better harness the power of AI-powered question-answering systems and create a more effective learning experience. The future of AI communication lies in our ability to ask the right questions and guide the response process, and this chart serves as an excellent foundation for that journey.
