🌚 Discover


Prompting generative pretrained transformers (GPTs) for effective assistance can be difficult. While 'prompt engineer' is a bit misleading in its title, there is real value in structuring the way you ask and communicate for ideas, text, or conversation when interacting with AI & LLMs. If you are using ChatGPT, you can leverage some of the techniques below with the power of the custom instructions option to empower your chat sessions with even more specific context built around your project or chat. These concepts can apply for any system message or prompt injection/chain technique.

Assistant Prompting Tips

To utilize emerging AI systems to their full potential, it's crucial to understand how to prompt efficiently. Here are some tips:

  1. Be explicit: Clearly state what you want the model to do. If you're ambiguous, the model might generate responses that are not as precise as you'd like. For example, instead of prompting, "Tell me about dogs," you could say, "Provide a brief overview of the history, characteristics, and common breeds of dogs."

  2. Specify the format: If you want the answer in a particular format, indicate it in your prompt. For instance, "Generate a bullet-point list of the top 5 tourist attractions in Paris."

  3. Ask for step-by-step instructions: If you want the model to guide you through a process, request it to provide step-by-step instructions. For example, "Provide step-by-step instructions on how to bake chocolate chip cookies."

  4. Use a system message: You can start with a system message to gently instruct the model about its role. For instance, you can start with, "You are an assistant that speaks like Shakespeare." The AI will then respond in a style resembling Shakespeare's.

  5. Temperature and max tokens: You can also adjust the 'temperature' and 'max tokens' parameters to influence the output. A lower temperature value (e.g., 0.2) makes the output more focused and deterministic, while a higher value (e.g., 0.8) makes it more diverse. The 'max tokens' parameter limits the length of the generated text.

Assistant Examples

Example 1: Programming Assistant

If you're learning a new programming language, you might need to ask for explanations or code examples. Here's a sample prompt:

"You are a programming tutor specializing in Python. I'm a beginner looking to understand how to create a simple 'Hello, World!' program. Could you please explain the steps in a clear and easy-to-understand manner?"

The AI assistant, given its context, would provide a detailed, beginner-friendly explanation and a simple Python code snippet. If the response is still too hard to understand, get creative! Using specialized approaches to frame topics can help you break down a complex topic in a relatable way. For example, you could try:

"Explain that last snippet using video game and RPG analogies that effectively educate me on the script line by line. Then, write me a summary of your notes at the bottom of your response recapping our script."

Adjusting your response as you go can lead to immersive ways to learn and engage with your content, making scripting fun in ways that make sense to you (and in your preferred style of learning).

Example 2: Project Assistant

Perhaps you're managing a software development project and you need advice on agile methodologies. You could ask:

"You are a project management expert with a focus on Agile methodologies. I'm leading a software development project for the first time and I'm a bit confused about how to implement Scrum effectively. Could you please provide a brief overview and best practices?"

Depending on its response, you could follow this up with something like:

"Now, take your last response and reshare after formatting it into an actionable contract I can provide to (insert profession here) covering all necessary details, actions, and prerequisities for accomplishing the task(s) within (x,y,z) parameters."

Embed these metaprompts as you go, adjusting to suit your workflow styles and preferences.

Example 3: Personal Fitness Assistant

If you have set a personal fitness goal, such as running a marathon, and need guidance on training plans, you might prompt:

"You are a fitness coach specializing in marathon training. I'm planning to run my first marathon in six months and I need a structured training plan, including nutrition and rest days. Can you help me with that?"

From here, you could follow up the response with:

"Now, take the advice you've given me so far and reshare it in a comprehensive meal plan format that is easy to read and easy to follow within these preferences: (x,y,z)"

Weave your chats to be contextual to your goal so that you're actively working towards collecting information and actionable steps that can be consolidated and formatted by the model at the end of the conversation. Utilize this chat format as a sort of advanced research mechanism that gets you started on any project, topic, or material you are engaging with.