[With Examples] How to Master Chat-GPT 07: Decoding the [26 Principles of Prompt Input] Part 6: The Prompt That Makes English Conversation Schools Unnecessary!?
The "26 Principles" for optimizing interaction with Large Language Models (LLMs) published by a group of researchers at the Mohamed bin Zayed University of Artificial Intelligence (MBZUAI). I have provided supplementary explanations up to principle 14 so far.
This time, we start from 15.
15. Instruction including testing: Have it give you a test to check your understanding -> △ Sometimes it's not great when trying to match your level of understanding ◎ English learning methods using Chat-GPT as an instructor are a must-see
There are two ways to use this prompt.
First, when you want Chat-GPT to explain something to you, you will get a more appropriate response if you tell Chat-GPT your level.
For example, when you want Chat-GPT to teach you about something, if you tell it in advance how much you already understand about it, it will return an answer at the "level" that is optimal for you.
You could say things like "Explain it to me like I'm an elementary school student" or "I have some knowledge, so answer with that in mind," but the most reliable way is to use a prompt like the one below.
[Example] When you want it to teach you according to your level of understanding
I want to know about XX.
I want you to teach me concretely, according to my level of understanding.
It is better for you to confirm how much I understand about XX before responding, so please proceed as follows.
1. Test me on how much I understand about XX.
2. I will answer the test.
3. Based on my answers, grasp how much I understand about XX.
4. Provide an appropriate explanation of XX to me based on the level of understanding you have grasped.
In my experience, the method of using this prompt is sometimes effective and sometimes not so great.
This is because, while it is fine to have test questions issued and to answer them, sometimes it provides explanations that are limited in scope when you answer.
[Example] A prompt that allows Chat-GPT to become your English instructor
Another way to use it is to have it create test questions based on what you learned from Chat-GPT to check how much you understood, and I felt that this is most effective for English conversation and English vocabulary practice.
For example, let's try having a free talk in English about a news article.
Copy and paste the news article and enter the prompt as follows (if you want a professional-level lesson, I think you should write a more elaborate prompt, but you can still have a pretty good lesson with this).
Me
・You are an English conversation instructor.
・My English conversation level is high school level.
・Please have a free talk with me in English based on the news article below #News.
・Understand the content of the article and give me 50 test questions to check if I understand the article.
・Please give the test questions one by one.
・After I answer the test question, please judge whether the answer is correct.
・If it is not correct, please output the reason and a correct example answer in English.
・When the test is finished, please evaluate my English conversation ability on a 5-point scale and list 5 points for improvement.
#News
・・・(Paste the news here)
Must-see! Turn it into an "English conversation teacher" by adding the smartphone voice function
If you want to have an English conversation, there is also a way to use the voice function of the smartphone version of Chat-GPT to interact via voice.
With this, Chat-GPT truly becomes your English conversation lesson teacher. I think you won't even need an English conversation school! Haha.
16. Assigning a role to the model: Assign a specific role to the model → 〇 Assigning a role leads to more 'authentic' responses
This has come up many times, hasn't it?
・You are an English conversation teacher.
・You are a management consultant...
By specifying a role at the beginning, you will get more 'authentic' responses back. I'm sure this is a prompt everyone has used at least once.
17. Using delimiters: Use specific separator characters → ◎ Clearer communication. Even just incorporating 'line breaks' helps
You might be wondering, what is a delimiter?
A delimiter is a computer term for a symbol or string used to indicate the separation or structure of data, commonly used in programming and data processing.
For example, commas in CSV files, tabs in text files, semicolons in programs, etc. These play an important role in programming and data processing to facilitate data splitting and analysis.
It is also effective for Chat-GPT. This is because when entering a prompt to give instructions to Chat-GPT, it is quite difficult for it to understand where the breaks are in the entered text.
The delimiters I often use are
'#', '{}', 'line breaks', '・', '-', and ':'.
[Example] A template using delimiters
Please look at a prompt template I wrote previously.
# Request
You are a [#Role].
Please answer the [#Request Items] in the format of [# Format].
Please execute according to the [#Procedure].
Please execute while following the [# Rules].
Please execute by referring to the [# Reference].
# Role
- A public relations representative for a manufacturer promoting a new product on social media
# Request Items
- Create a product catchphrase that is effective for advertising the new product, around 15 characters long.
- Write a product advertisement text based on the catchphrase, around 200 characters long.
- Create 5 sets of catchphrases and advertisement texts.
# Format
- Use a table format
- Columns are [Draft Number] [Product Catchphrase] [Product Advertisement Text]
- Rows are 5 rows. Display 5 sets of [Product Catchphrase] and [Product Advertisement Text]
# Procedure
1. Provide an answer. Once you have answered, ask me, 'Is it okay to finish the answer?'
2. If I input YES, end the execution.
3. If I input NO, ask me, 'Which draft do you like?'
4. Create 5 new drafts similar to the draft number I answered.
5. Return to step 1 of the procedure. However, automatically end after executing the #Procedure 10 times.
# Rules
- Write casually to make the product feel approachable
- Write catchphrases and text that will be accepted by young people
# Reference
Paste the new product information text here
Many delimiters are used, aren't they? By using these, you can convey the breaks between words and meanings to Chat-GPT more clearly.
[Important] 'Line breaks' are key when using delimiters
By the way, I think there are still few people who use delimiters. It might be troublesome at times, but in such cases, please use at least one.
The key here is 'line breaks'.
Just by breaking up the prompt and inserting line breaks, the quality of the answer will improve significantly.
Specifically,
Before
"You are an English conversation teacher. Please conduct an English conversation lesson based on the article I have pasted, using free talk. After 20 exchanges, please ask me questions to gauge my level of understanding. The content of the news article is..."
Rather than that,
After
#{Role}: English conversation teacher. (Line break)
Please conduct an English conversation lesson according to the #{Procedure}.
# {Procedure}
・Please read the #{Article}.
・Please engage in free talk regarding the content of the read #{Article}.
・Please converse with me for 20 exchanges.
・Please lead the conversation.
・Please actively ask questions and bring up topics.
・After 20 conversation exchanges are finished, please ask me questions about the #{Article}.
・Please ask questions to confirm how well I understand the #{Article}.
#{Article}
XXXXXXXXXXXXX...
And so on, please insert line breaks and input the instruction prompt in segments.
Just by doing that, Chat-GPT's comprehension improves significantly.
I also asked Chat-GPT!
In fact,
I
Is using delimiters in prompts effective for you?
GPT
Using delimiters in prompts can be effective for ChatGPT. By separating different elements or instructions within a prompt, you can facilitate clearer understanding and more appropriate responses. For example, by separating different questions or topics with line breaks or specific symbols, it becomes easier for ChatGPT to recognize each part as an individual element.
That is what it said!
To be continued next time.
"26 Principles" for Optimizing Interaction with Large Language Models (LLM)
① Omit politeness: Polite terms like "please" are unnecessary
② Specify the target audience: Specify who the response is for, such as "for experts"
③ Break down complex tasks: Divide complex tasks into simple prompts
④ Use affirmative instructions: Use positive language rather than negative words
⑤ Instructions for clarification: "In simple terms," "So that even a middle school student can understand," etc.
⑥ Offer rewards: Indicate that "a reward will be given for a good answer"
⑦ Provide examples: Use existing examples
⑧ Prompt format: Start with "###Instruction###" and include "###Example###" or "###Question###" as appropriate.
⑨ Clear task instructions: Instruct with "Your task is".
⑩ Indicate penalties: State that "there is a penalty".
⑪ Instruction for natural language response: Instruct to "answer in natural language".
⑫ Use leading words: Instruct to "think step by step".
⑬ Eliminate bias: Instruct to "be unbiased and not rely on stereotypes".
⑭ Promote dialogue with the user: Have the model ask questions until the problem is solved.
⑮ Instruction including tests: Have the model give you a test to check your understanding.
⑯ Assign a role to the model: Assign a specific role to the model.
⑰ Use delimiters: Use specific separator characters.
⑱ Use repetition: Use specific words or phrases multiple times.
⑲ Chain of thought: Generate intermediate steps and combine examples.
⑳ Use output primers: End the prompt with the beginning of the expected output.
㉑ Instruction for detailed text creation: Instruct to "write in detail".
㉒ Prevent style changes: Instruct to "not change the style".
㉓ Coding prompts for multiple files: To streamline coding tasks spanning multiple files, suggest creating a script that automatically creates new files and inserts the generated code into the appropriate files.
㉔ Continue with specific words: Instruct to "complete with the given words".
㉕ Specify model requirements: Clearly state the requirements the model must follow to produce content in the form of keywords, regulations, hints, instructions, etc.
㉖ Text creation based on samples: Instruct to write in the same language as the sample.