Introduction
When training an AI model, providing it with the right data is crucial. However, the way you prompt the model during this process can significantly impact the quality of the output. Prompts serve as the instructions that guide the AI on how to interpret and generate responses based on the data it's been trained on. Even the best-trained model can produce subpar results if the prompts are unclear or poorly structured. This guide will help you understand how to create effective prompts, providing examples and use cases to ensure your AI models perform at their best.
Training
Starting a new training session in Dojo AI is simple. Here’s how you do it:
Click on the "Training" Tab: This is where you can keep each unique dataset.
Click on the "New Dataset" Button: This begins the process of creating a new dataset for your AI model.
Choose the Type of Content: You can select from various options like Tweets, Emails, LinkedIn posts, Short Form Scripts, or "Other" for custom types.
Use Templates for Easy Prompting: Templates help you by setting up the prompts automatically. For example:
If you choose LinkedIn, the prompt template will be: “Write a LinkedIn Post about %TOPIC% in the style of LinkedIn.”
You only need to fill in the %TOPIC%. So, if your topic is "pizza," the AI will understand it as: “Write a LinkedIn Post about pizza in the style of LinkedIn.”
System Prompt: This is the overall instruction for the model. If you choose LinkedIn, the system prompt will be: “You are an expert LinkedIn ghostwriter. I am going to give you a topic. Your only job is to craft a perfectly worded post, in the style of LinkedIn.”
If you select "Other," you'll need to create these prompts yourself.
Understanding Prompt Components
Prompts are essential in guiding AI models to generate desired outputs. When training, there are three main prompts needed for the AI model to effectively understand your specific task. The overall instructions, the topic you provide it, and the desired output you want the trained AI model to sound like.
In Dojo AI, these prompts are structured into four main components: System, Prompt Template, Input, and Output.
System: The overall instructions for the AI model
Prompt Template: This is what you provide to train the AI model.
Input: This is the variable %TOPIC% that you specify in the prompt template.
Output: This is the desired response example you want generated by the trained AI model.
Example:
System: You are an expert at writing effective LinkedIn posts. Please write a LinkedIn post about the topic I provide.
Prompt Template: “Write a LinkedIn Post about %TOPIC% in the style of LinkedIn.”
Input: “Remote Work is better”
Output: “Remote work is for geniuses.
It's revolutionized the way we approach our careers.
It offers flexibility, increases productivity, and provides a better work-life balance.
I don't care what anyone says, my company will forever be remote.
Embrace it, boomers.. 👊”
Here, the AI is trained on the full context, but all you need to do is input the topic along with your example to teach the AI model your exact style, tone, and format. Notice the unique tone, spacing, use of emojis, the model will learn from all of this, and closely follow your examples.
Creating Effective Prompts
To create effective prompts, it’s crucial to be clear, concise, and specific. Here are some guidelines:
Clarity: Ensure the prompt is easy to understand.
Specificity: Define the exact output you expect.
Context: Provide enough context to guide the AI model.
Examples of Good Prompts
Here are some examples of well-crafted prompts for different content types:
Twitter Posts
System Prompt: “You are an expert Twitter ghostwriter. I am going to give you a topic. Your only job is to craft a perfectly worded tweet, in the style of Money Twitter.”
Prompt Template: “Write a tweet about %TOPIC% in the style of Money Twitter.”
Input: “Pizza”
Output: “pizza is the best food on the planet and i’m ready to die on this hill”
Notice how we keep it all in lowercase, as that is the desired content we want to generate. Not every single example we give the model will be lowercase, but it helps if you want a variety of outputs including styles like this.
LinkedIn Posts
System Prompt: “You are an expert at writing effective LinkedIn posts. Please write a LinkedIn post about the topic I provide.”
Prompt Template: “Write a LinkedIn Post about %TOPIC% in the style of LinkedIn.”
Input: “Personal branding and the importance of content”
Output: "Personal branding is the future.
It doesn't matter what you do. It doesn't matter where you work.
If you're good at what you do, people should know who you are.
And the best way to do that is by creating content.
Here's why:
- It builds trust with potential clients.
- It makes you the go-to person in your industry.
- It's the best way to scale your business.
If you're not creating content, start today.
You won't regret it."
Notice how we train it with the spacing like this. This is because we want this as our style consistently on LinkedIn, and it's much easier to take that content or slightly modify it for our needs.
Short Form Video Hooks
System Prompt: “You are an expert at generating viral hooks for short form video scripts. Respond with the hook only, no other commentary.”
Prompt Template: “Create a video hook about %TOPIC%.”
Input: “AI replacing copywriters”
Output: “If you’re a copywriter, you should be worried about this
AI is getting so good that it can now replace copywriters with a click of a button”
Conclusion
Effective prompting is key to leveraging the full potential of Dojo AI for generating high-quality content. By understanding and utilizing well-crafted prompts, you can guide the AI to produce outputs that are relevant, engaging, and aligned with your specific needs.
Related Resources
Community: Access support from experts inside our Discord Community
Additional Guides: Check out some additional guides on our blog.