Overcoming The Limitations Of ChatGPT: These 2 Things Are Key To Getting The Best Results

Many people think that getting the most out of ChatGPT is all about the prompts we feed it. Whilst the prompt is important, there’s another factor at play here that is often overlooked: effective priming.

It’s not uncommon for people to confuse the two, so allow us to explain the difference:

  1. Prompting: the textual inputs (such as questions, phrases, keywords discussion topics, or instructions) that you input into ChatGPT in order to generate a response. 

  2. Priming: a method of providing context to ChatGPT to improve responses, enhance its understanding of a prompt, fill knowledge gaps, and train it to meet specific needs.

Let’s explore each and why they’re important.

What is priming?

Knowing how to prime ChatGPT is critical to getting the most out of it. However, many people often get it wrong by providing a lack of context and ignoring tone and style.

If we do not give ChatGPT adequate context, it will struggle to know exactly what we want. As we mentioned in the previous blog, ChatGPT is trained on patterns. Consider for example asking ChatGPT to write a blog for us. If we provided very little context, it might answer our instruction, but the blog it produced would be based on the patterns and training it knows – not patterns unique to us. 

It would not be specific to our audience, written in our tone, or formatted in the way we might prefer. This is a common occurrence with people using ChatGPT incorrectly. It’s an ineffective approach because it means we have to rewrite or edit the text to sound more like us, which ends up negating the time-saving advantage ChatGPT should offer.

How can we effectively prime ChatGPT to garner better results?

Much like in a theatre production where there are sets, props, and a whole lot of lighting and sound technicalities, ChatGPT also requires setup and additions to run smoothly.

The most important thing to be aware of is customisation. Yes, we can customise ChatGPT to work for our unique business needs, tone, and style! But first, it needs to be fed context about who we are as a business. 

By priming, we can customise our experience so that the AI better understands who we are, what our business does, and who we help – meaning everything we generate is tailored uniquely to us. 

Taking the time to carry out this preparatory step dramatically improves the responses we’re likely to receive. To do this, we can feed ChatGPT specifics about our business. For example, a step-by-step description of our customer journey, specific formats and frameworks unique to our business, and information about the nuances of not only our industry but also the products and services we offer as a business.

Where priming and prompting meet

When using ChatGPT to produce marketing or written copy, we can prime it by sharing examples of our own writing. This provides the AI with context about our tone, language, and writing style. Doing so allows us to train it on patterns that are unique to us, meaning the responses we generate will sound more like they were written in our voice. 

Here’s an example of a prompt, including a prime, written by ChatGPT expert and founder of AI Edge, Tiffany Bridges:

“To better optimise the understanding of my brand voice, writing style, vocabulary, level of readability, and tone, I would like to request a comprehensive analysis. 

Taking into consideration the provided writing sample, could you dissect and distil its key elements in a manner that allows for a clear and coherent understanding? I aim to utilise your insights to emulate my unique writing style in your generated responses. Please, detail this in a format that will best facilitate your imitation of my writing.” 

This prompt works by sending a piece of our own writing first (the prime), and then following with the above instruction. Within this prompt, we can see reference to the prime: “taking into consideration the provided writing style”, which gives ChatGPT the context it needs to produce the best result. 

How can we write effective prompts for ChatGPT?

Many of the common mistakes people make with priming are also seen in prompting. The prompts might be too vague, offer little context, or fail to provide enough information about tone, style, voice, or format.

The biggest tip we can give is this: we need to be constantly iterating and learning. As much as we are training ChatGPT to produce accurate and high-quality responses unique to us, we also need to be trained in providing accurate and high-quality instructions.

So, what are some ways we can craft expert prompts?

  1. Begin with the end in mind.
What do you want from it? What is your ideal end goal?

2. Be specific. Give it a clear ask.

Are you being extremely clear? What format do you want it in?

3. Give it context.

We can’t overstress the importance of first priming the AI before we give it an instruction. A great way to do this is by providing it with an example of what you’re looking for or what pattern it should follow.

4. Feed it tone and style

If we want to produce content that sounds like us and matches our unique tone and style, we need to be allowing ChatGPT to “get to know us” by providing it with examples.

5. Iterate 

ChatGPT is constantly developing, and so is our understanding of how to get the best results from it. A great way of improving our approach is to reverse engineer it. If ChatGPT provides an inaccurate or disappointing response, we can ask it, “What prompt should I have used to obtain the result I wanted?” 

Understanding the limitations of ChatGPT in our prompts 

When designing prompts for ChatGPT, it’s important to bear in mind that whilst it can handle quite an extensive list of things to take into consideration, it struggles with overly complex instructions.

For example, if we simply asked ChatGPT to write a book for us, it is highly likely that it would produce an inaccurate response. However, if we broke it down and first asked for a framework, format, and outline, and then took it chapter by chapter, it would have a much better chance – especially if we primed it properly beforehand.

This is a good example of why iterative or sequential prompting is so important when you’re trying to achieve complex tasks using ChatGPT. We can’t always expect to feed it a “God prompt” – that most perfect prompt that generates exactly what we’re looking for straight away and with no edits.

Getting the best results from this fantastic technology requires not only learning and development on ChatGPT’s part, but on our end too. 

P.S. If you missed our previous blog post, 3 Ways You’re Using ChatGPT Wrong & How To Harness Its Full Potential, which offers an introduction to ChatGPT, its use cases, and its limitations, check it out here.

If you’re interested in implementing automated or AI technology into your business, it couldn’t be easier to hop on a free discovery call with The Business Catalyst to see how we can help. 

For more information on Tiffany, visit AI Edge here or follow along on LinkedIn

The full webinar is available here