Easy Way To Increase The Quality Of Your Prompts

Hello.

In today's article, I want to show you an easy way to improve the quality of your prompts.

This technique shouldn't take more than 2~3 minutes to implement, and it baffles me that more people aren't using it.

So if you're crafting prompts that need to be highly reliable and with minimal errors, this technique is for you.

But before we dive into the technique itself, let's first discuss why it's necessary. I promise I won't bore you with technical jargon, but it's important to understand a few things.

You see, Large language models are unreliable for a few reasons...

• They have a knowledge cutoff. Which means their information is not up to date.

• You are never sure if the model is pulling information from sources you trust.

• If the model doesn't know an answer, it might give you convincing nonsense instead of admitting it doesn't know.

Needless to say, this is a significant issue for those who need their prompts to be highly reliable.

And that's exactly where the prompting technique I am about to show you comes into play...

Overcoming LLM Faulty Memory

Retrieval Augmented Generation, or RAG for short, is a method that helps you overcome GPT's limited memory.

With RAG, you'll be able to give the model more context and information from external sources, leading to more accurate answers and fewer errors.

And despite its technical-sounding name, it's easy to use.

All you have to do is upload a PDF file with all of the information that you want it to have access to.

And after you've uploaded the file containing all the information you want GPT to know, the model can use it to answer your questions.

Let me give you an example of how this works, and it will become clearer to you...

RAG In Action

Take a look at the following GPT assistant.

Its goal is to help book readers create prompts based on books they have just read.

Of course, people can try doing it by prompting GPT directly, but they would inevitably face one of the hurdles we talked about earlier:

• The model will probably hallucinate some results since it does not have access to the specific information you are asking it for.

• The result will be so bland and generic as to be almost useless.

So, if you want more reliable and actually useful results for this particular task, you would need to use RAG.

1. Find a PDF Document With The Information You Want.

The first step is to choose a good document that will provide context for the RAG.

Ideally, this document must have every piece of information that you want GPT to have access to.

For this specific example, the ideal one would be an e-book PDF.

However, depending on your needs, you can use:

• Online articles;

• Research papers;

• Company Reports;

• Literally any material you can find that you think can help GPT generate better and more accurate answers.

Anything that you can upload to GPT will work. Use yourcreativity to see what fits best for your needs.

There's always something you can give to it to get better outputs.

2. Proceed as normal.

The second step is to prompt GPT as you normally would. Since you've already uploaded the document, GPT now has access to all of the information from that file.

What To Use As Context For The RAG

One point I'd like to stress is that there's always a document you can upload to enhance the output quality for your specific needs.

It could be a research paper, a web article, or even notes from your note-taking app.

Be creative.

It's always worth the effort to think of a relevant source to use, especially if your use case requires highly accurate results.

PS: If you want to learn how to create world-class prompts, I have a guide where I teach you exactly that.

There are plenty of lessons on cutting-edge prompting techniques, as well as examples and breakdowns of prompts from great products.

On top of that, I am always updating the guide adding the newest prompting techniques as they come up.

Check out the guide here.