If you use AI tools as much as I do, you're probably already pretty good at asking AI for what you need — also known as “prompting.” But there’s always more to learn. Using advanced prompting techniques can help you unlock some of the best features of large language models (LLMs) like ChatGPT. 

Thanks to new research from a team at Vanderbilt University, we've collected a few of the most useful advanced prompting techniques you can use to get better results faster, no matter what you're using ChatGPT to do.

1. The Persona Pattern

Prompt pattern:
From now on, act as [persona]. Pay close attention to [details to focus on]. Provide outputs that [persona] would regarding the input.

Sure ChatGPT is useful, but its outputs can also be a bit generic. The Persona Pattern aims to alleviate this issue and instead allow the LLM to adopt a specific point of view. After that, the LLM can interact with you from this different perspective. 

The Persona Pattern can be adapted to a number of different useful scenarios. For instance, fiction authors can use it to chat with their characters for a deeper understanding. Non-fiction writers can use it to empathize with their readership and gauge what might resonate with them. I've found that it's particularly useful as a brainstorming partner — one that doesn't get tired of answering questions.

Another useful way to use the Persona Pattern is to ask ChatGPT to act as an expert that can give you advice. For instance, you might ask it to help coach you on your writing to see how you can improve. The beauty of this approach is that it can give you customized feedback instead of a checklist of generic pointers.

2. The Flipped Interaction Pattern

Prompt pattern:
From now on, I would like you to ask me questions to [do a specific task]. When you have enough information to [do the task], create [output you want].

Typically, you pose questions to the language model, which in turn generates an output. The Flipped Interaction Pattern turns this dynamic on its head, with the LLM asking questions for you to answer. 

Why take on the role of the answerer? There are two main advantages, one for the LLM and the other for you. For the LLM, this prompting pattern allows it to gather the information it needs to answer the question you've given it in a more precise and tailored way. Otherwise, you might think that you’re giving all the required information only to find that the LLM is actually missing critical details that you overlooked. It can also highlight times where it needs information you don't yet know or haven't thought of and prompt you into finding it. 

The other advantage is that it makes the interaction easier on you. Instead of investing a ton of time considering your task and crafting a long prompt with all the details you think it might need, the AI simplifies the process by posing a series of questions that you can answer in bullet points or paragraph format. 

I found that it was a good exercise to make me think more deeply about the context and specifics of what I wanted instead of triggering a frustrating back and forth with the tool. In other words: you get to the desired result faster.

‍

‍

3. The Question Refinement Pattern

Prompt pattern:
From now on, when I ask a question, suggest a better version of the question to use that incorporates information specific to [use case] and ask me if I would like to use your question instead.

We're often unaware of additional information that might be helpful when crafting a prompt, especially if we’re not experts on the subject. Enter the Question Refinement Pattern, which allows the LLM to refine your initial prompt by suggesting an improved version of your question.

Instead of prompting directly, you ask the LLM to offer an alternative question that includes specific information relevant to your use case every time you ask a question. The LLM then asks you if you want to use that alternative question instead.

This pattern works best in areas where your initial question is broad or vague and would benefit from letting the AI tool guide you through what kinds of context is important to add. Think about how much technical knowledge about the topic might help you and if you're unsure about the specifics that are relevant, it's a good candidate for this pattern.

In cases where additional context has less importance or when the situation is highly variable and more context won't help refine the answer, the results are not as impressive. And that's not always as easy as it sounds. I found it difficult to differentiate between these two, so my results using this pattern were mixed – ironically, the best performance I got with this prompt pattern is when I asked about the prompt pattern itself.

Here’s an example of a time you might use this prompt pattern:

User:

From now on, when I ask a question, suggest a better version of the question to use that incorporates information specific to writing a script for a top-quality podcast and ask me if I would like to use your question instead.

ChatGPT:

Of course! I can certainly help with that. Just to make sure I fully understand your request, can you provide a sample question you might ask? This will allow me to provide a more accurate revision suggestion for future questions related to scriptwriting for a high-quality podcast.

User:

Can you write a podcast script based on the following article which was in the June 20, 1923 Sunday edition of the New York Times.

FRENCH PEOPLE QUARREL OVER JOAN'S APOSTROPHE

Royalists Spell Her Name "d'Arc," While Peasant Partisans Simplify It to "Darc"

[article]

‍ChatGPT:

Sure. Here is a revised version of your request, focused on creating a compelling podcast script:

"I'm preparing to produce an engaging episode of our history podcast, and I found a fascinating historical dispute from a New York Times article published on June 20, 1923. It discusses the controversy surrounding the spelling of Joan of Arc's name in French - "d'Arc" versus "Darc."

Can you help me write a captivating podcast script that intricately weaves this story? I want to highlight the socio-political implications behind each spelling choice and the violent clashes that erupted due to this seemingly trivial difference. Also, emphasize the diverse characterizations of Joan by French citizens, from a noble heroine to a simple peasant girl.

Remember, our audience enjoys rich historical narratives, an injection of humor where appropriate, and an accessible breakdown of complex issues. How does that sound?"

‍

4. The Cognitive Verifier Pattern

Prompt pattern:
When I ask you a question, generate three additional questions that would help you give a more accurate answer. When I have answered the three questions, combine the answers to produce the final answers to my original question.

The Cognitive Verifier Pattern, much like the Flipped Interaction Pattern, asks the language model to generate questions that you answer to help it provide a better response. Like the Flipped Interaction Pattern, the final part of the prompt makes sure that the language model incorporates all the component parts into the final result.

In this pattern, though, speed isn't the main objective. Instead, it's about the LLM's reasoning process. When you deliberately ask the language model to break the question down into smaller component parts, its reasoning often improves — so it's a good pattern to use on more complex tasks. By going step by step, it can also identify parts that are missing or where the prompt was unclear. 

This pattern is most useful when you're asking the LLM to help you complete a task that requires a logical sequence. Organizing narratives and outlining longer-form content are good candidates for this pattern. 

When I used it to plan a podcast episode, it helped me consider the balance between two conflicting viewpoints — a more sophisticated question than those generated using the Flipped Interaction Pattern. I also found that this pattern can give better results when brainstorming possible scenarios or directions for a narrative, like trying to work out a realistic plot or effective strategic plan.

5. The Reflection Pattern

Prompt pattern: 
When you provide an answer, please explain the reasoning and assumptions behind your response. If possible, use specific examples or evidence to support your answer of why [prompt topic] is the best. Moreover, please address any potential ambiguities or limitations in your answer, in order to provide a more complete and accurate response.

With a more basic prompt, you might overlook the reasoning behind certain parts of it. The Reflection Pattern makes this reasoning explicit and gives insight into the reasons why the LLM makes the suggestions that it does. 

There's an added bonus too: the Reflection Pattern generally provides better reasoning on the part of the language model. The very process of asking the LLM to provide its reasoning has been shown to make it perform better. Win-win.

The Reflection Pattern is most effective when you're asking for something that will be challenging for you to judge. For example, if you're asking ChatGPT to turn an article into a podcast, but you've never produced a podcast before, you can use this to see why it's making certain suggestions.

I found this prompt pattern to be pretty insightful when I was asking for things I had less experience with. It explained why it included certain elements in its output, some of which pleasantly surprised me. I also have had good success with getting the reasoning behind its edits to my writing, giving me more context on the suggestions and making it easier to decide which ones to accept or reject.

Conclusion

These advanced prompting techniques unlock the full potential of AI tools like ChatGPT. They can enhance the tool's precision and also provide you with a more intuitive experience. Each method has unique strengths that cater to different use cases, so don't hesitate to experiment and discover which pattern yields the best results for you.