Creating a style guide to use in GPT Prompts

As discussed in my previous article on LLM Prompt Engineering, keys to getting better results from GPT models is to be specific and to provide contexts. That might seem a bit overkill if you just want it to write some paragraphs of text, or a poem, or a haiku, but don’t you want it to write something that sounds like it came from you? You don’t just want it to sound like everyone else!

One of the ways that you can do this is to provide a style guide in the prompt. These can be quite detailed and specific, so how do you go about crafting one of your own? Well, ChatGPT can help!

First, grab a piece of text that you have written of a reasonable length - blog post rather than War and Peace though! Then just paste it straight into the ChatGPT chat window.

For the purpose of the demonstration, I used the text from my article called what is the real experience of analytics in the metaverse. When I pasted the article into the window, the default action is to summarise it. The next thing I asked ChatGPT to do was, “Analyse the style of writing in the article.”

The response was a quite detailed (and very flattering!) analysis of my writing style. I followed up with the request, “Write a GPT Prompt style guide for this author’s style”.

It used the 6 bullets to create a detailed style guide. I did feel that it was quite wordy though and asked how many tokens that was - 236 was the reply. I then asked, “Could the style guide be written more concisely to achieve the same outcomes?” This resulted in a new guide that was 96 tokens - a good improvement if I was tight on tokens!

Now, I have experiment a little more with it, including asking to add a little “Hiberno-English flair”! I have used it to create the article below, which I have not edited.

I would be interested to hear what you think! Does it sound like me?

Title: Advanced Prompt Engineering Techniques for GPT-4: Unleashing the AI's Potential

Introduction: The Power of Prompt Engineering

With the rise of GPT-4, harnessing its potential through advanced prompt engineering has become essential. As an AI enthusiast, I've been exploring ways to make GPT-4 more efficient and effective. In this article, I'll share techniques to help you craft powerful prompts for this advanced language model.

1. Crafting Precise Queries

GPT-4's accuracy depends on the clarity of your query. To get better results:

• Specify the format you want the answer in.
• Provide context or examples for complex concepts.
• Avoid ambiguous terms or phrases.

2. Step-by-Step Instructions

Breaking down tasks into step-by-step instructions can guide GPT-4's output more effectively. This method:

• Reduces the chances of incomplete or incorrect responses.
• Allows GPT-4 to focus on one aspect at a time.

3. The Use of Constraints

Adding constraints to your prompt can limit GPT-4's scope, producing more targeted responses:

• Set word limits or response formats.
• Use phrases like "in three sentences" or "list three reasons."

4. Leveraging External Knowledge

GPT-4 may not have the latest data, so providing updated information can improve results:

• Offer relevant facts, statistics, or expert opinions.
• Encourage GPT-4 to use the provided data in its response.

5. Iterative Prompting

GPT-4's responses can be refined through iterative prompting:

• Use the model's output as a starting point for a new prompt.
• Ask for clarification, expansion, or rephrasing.

6. Experiment and Iterate

The key to mastering GPT-4 prompt engineering is experimentation:

• Test various prompts and analyze the outcomes.
• Modify prompts based on feedback and results.

Conclusion: Unlocking GPT-4's Potential

Advanced prompt engineering techniques can vastly improve GPT-4's performance. By crafting precise queries, breaking tasks into steps, adding constraints, leveraging external knowledge, and iterating, you can unlock the true potential of this powerful AI language model. So, dive in, experiment, and discover the possibilities!