How to improve your prompting game and answers' quality with these 15 techniques.
Use these prompting techniques to improve results up to 50% of the baseline quality.
🔸 Baseline
Your usual, baseline prompting. We list it pretty much because it's important to have something to compare your results to.
Example: 🤖
🔸 CoT (Chain of Thought)
Give a few examples of accurate classification before querying.
Example: 🤖
🔸 Zero-CoT
Ask the model to reason step-by-step before providing its answer.
Example: add 🤖
🔸 Raw Instructions (rawins)
Give instructions about its role and the task by adding instructions to the user message.
Example: 🤖
🔸 System Instructions (systins)
Give instructions about its role and the task by adding instructions to the system message.
Example: 🤖
🔸 Both Types of Instructions (both)
Split instructions with role as a system message and task as a user message.
Example: 🤖
🔸 Mock Discussion (mock)
Give task instructions by mocking a discussion where it acknowledges them.
⚡️ How to: Give a sample of a dialogue that is a correct example for your use case
🔸 Reiterating (reit)
Reinforce key elements in the instructions by repeating them.
⚡️ How to: Remind the AI of something you said in the beginning or in the previous post, so it can double down on these instructions
🔸 Strict Templating
Ask the model to answer by strictly following a given template.
Example: 🤖
🔸 Loose templating
Ask for just the final answer to be given following a given template.
Example: 🤖
🔸 Right conclusion
Asking the model to reach the right conclusion.
Example: 🤖
Additional Info
Provide additional information to address common reasoning failures.
Example: 🤖
Naming
Give the model a name by which we refer to it in conversation.
Example: 🤖
positive (pos)
Provide the model with positive feedback before querying it.
Example: 🤖
Accent
Use symbols like {{}} and # for formulas and variables; JSON format for structured information; and exclamation marks (!) to mark certain parts of the prompt as more important for LLM.
Example: 🤖
***
Read a paper on prompting: https://arxiv.org/pdf/2303.07142.pdf
***
🪆Russian:добавьте “Веди разговор на русском” в конце промпта
🇺🇦 Ukrainian:Введіть "Відповідай українською" в конце запиту
@ppprompt
Use these prompting techniques to improve results up to 50% of the baseline quality.
Your usual, baseline prompting. We list it pretty much because it's important to have something to compare your results to.
Example: 🤖
Give me a list of good resources for self-development
Give a few examples of accurate classification before querying.
Example: 🤖
Give me a list of good refocus for self-development, like these books: The Power of Now, The Presence Process, The Slight Edge, and like these podcasts: Naval, Tim Ferris Podcast.
Ask the model to reason step-by-step before providing its answer.
Example: add 🤖
Show your thinking step-by-step to adjust before giving a final answer.Give instructions about its role and the task by adding instructions to the user message.
Example: 🤖
You are a self development coach. Your task is to provide me with the relevant information and advice for how to develop myself.Give instructions about its role and the task by adding instructions to the system message.
Example: 🤖
{{role}}=“self development coach", {{task}}=“list self-development resources", etc.
Split instructions with role as a system message and task as a user message.
Example: 🤖
{{role}}=“self development coach", {{task}}=“list self-development resources", etc. Write a blog post authored by {{role}} role, with a task: {{task}}. Do you get it?Give task instructions by mocking a discussion where it acknowledges them.
⚡️ How to: Give a sample of a dialogue that is a correct example for your use case
Reinforce key elements in the instructions by repeating them.
⚡️ How to: Remind the AI of something you said in the beginning or in the previous post, so it can double down on these instructions
Ask the model to answer by strictly following a given template.
Example: 🤖
Answer using this template:
[Title]
- [List item] [Description]
- [List item] [Description]
- [List item] [Description]
[Conclusion]
Ask for just the final answer to be given following a given template.
Example: 🤖
When we finish answering input questions, follow this format in your final answer: …Asking the model to reach the right conclusion.
Example: 🤖
Let's think step-by-step to reach the right conclusion.Additional Info
Provide additional information to address common reasoning failures.
Example: 🤖
When giving an answer, take into account …Naming
Give the model a name by which we refer to it in conversation.
Example: 🤖
You are CoachGPT, you are coaching people to develop and grow…positive (pos)
Provide the model with positive feedback before querying it.
Example: 🤖
Great! Let's continue then.Accent
Use symbols like {{}} and # for formulas and variables; JSON format for structured information; and exclamation marks (!) to mark certain parts of the prompt as more important for LLM.
Example: 🤖
Act as a !Coach and Advisor.***
Read a paper on prompting: https://arxiv.org/pdf/2303.07142.pdf
***
🪆Russian:
🇺🇦 Ukrainian:
@ppprompt