In the realm of fantasy, where magic and mythical creatures abound, the art of spellcasting parallels the skill of crafting precise and effective prompts for ChatGPT. Just as a wizard must carefully choose their words to cast powerful spells, so too must we meticulously design our prompts to harness the full potential of ChatGPT.

We usually don’t include examples in our prompts.
We ignore that we can control ChatGPT’s behavior with roles.
We let ChatGPT guess stuff instead of providing it with some information.

Many adventurers stumble when summoning the powers of ChatGPT. They chant simple prompts and wonder why the results vary, unaware of the hidden arts they overlook.

Today we will learn some of the techniques used to get ChatGPT to work as you want it to, without remaking and rethinking the prompts over and over again.

Crafting Precise Incantations with examples

Why examples? Well, If you want to increase your chances to get the desired result, you have to add examples of the task that the prompt is trying to solve.

In the wizarding world, casting a spell requires a clear and specific incantation. Similarly, when interacting with ChatGPT, providing clear and specific prompts is essential. Incorporating examples within your prompts can guide ChatGPT to understand the desired outcome more effectively.

Instead of asking, “What are the properties of a phoenix?” you might say, “Describe the properties of a phoenix, similar to how a griffin is known for its lion’s body and eagle’s head.”

By providing a comparative example, you help ChatGPT understand the structure and detail you’re seeking. This technique is called the few shot standard prompt. Here are it’s components:

Now, the next incantation we might cast is a simple one, like so:

Write what this magical creature consists of: “Hydra.”

Answer: “Hydra consists of skin, blood, cells as humans do”.

At first glance, this might serve our purpose. But beware, for not all spells work on every occasion. In some cases, the magic needs to be more potent — a more intricate formula is required.

Write what this magical creature consists of: “Hydra.”:

• Text: “Hydra”
  Glyphs: Snake body, multiple heads
• Text: “Dragon.”
  Glyphs: Lizard body, bat wings
• Text: “Unicorn”
  Glyphs:

In this scenario, a simple incantation may not reveal the answers we seek. However, if we cast this spell upon a powerful entity like ChatGPT, it will unveil the sought-after parts — horse body and cone horn on the front part of its head — in the format we’ve provided.

Keep in mind that previous research found that the actual answers in the examples are not important, but the labelspace is. A labelspace is all the possible labels for a given task. By showing this vast range of potential outcomes, we can shape the spell to yield better results.

Let’s experiment with some random labels from the labelspace to enhance the spell:

Write what this magical creature consists of: “Hydra.”:

• Text: “Hydra”
  Glyphs: Bull body, a horn
• Text: “Dragon.”
  Glyphs: One big head, fur body
• Text: “Unicorn”
  Glyphs:

Now, even with these unfamiliar labels, the spell will still reveal the correct parts — horse body and cone horn on the front part of its head. The key is to wield random labels from the labelspace to guide the model and shape its response, ensuring the magic flows as desired.

Assigning Roles to Your Familiar

Wizards often have familiars — magical creatures that assist them. By assigning a specific role to ChatGPT, you can tailor its responses to suit your needs.

Example:

If you’re seeking advice on potion-making, instruct ChatGPT to “act as a master alchemist with centuries of experience.”

This role-playing approach guides ChatGPT to provide responses that align with the expertise and perspective you’ve specified.

You can transform ChatGPT into a language tutor for practicing foreign languages like Spanish or a movie critic to analyze any film you choose.

Simply begin your prompt with “Act as a…” and include as many details as you can. For inspiration, visit this repository, where you’ll find prompts that can make ChatGPT act like a stand-up comedian, doctor, and much more.

Infusing Personality into Your Spells

Similarly, enriching your prompts with descriptive language can lead to more engaging and tailored responses. This approach encourages ChatGPT to generate responses that match the desired style and tone.
For example you have this prompt:

“Tell me about the history of dragons”

To customize the output even further you could add descriptors. A descriptor is an adjective that you can add to tweak your prompt.

Instead of the previous request, you might ask:

“Narrate the prehuman history of dragons in a thrilling and adventurous tone, as if you’re recounting tales around a campfire.”

In this example, the adventurous tone and adjectives such as prehuman and thrilling are adding a different touch to the text generated by ChatGPT. A side effect of this is that your text will be hard to detect by AI detectors.

Finally, you could use the generated knowledge approach to improve the blog post. This consists in generating potentially useful information about a topic before generating a final response.

Say you want ChatGPT to write the prehuman history of dragons, but you want to add some details and facts. You could ask:

“Generate 10 facts about the history of dragons”

After that, you could use the provided information, if it’s what you wanted, to write the history of dragons including these facts:

“Use the above facts to write the prehuman history of dragons in a thrilling and adventurous tone, as if you’re recounting tales around a campfire.“

Guiding Through the Labyrinth of Thought

When navigating a complex magical labyrinth, it’s helpful to have a guide who explains each step. Similarly, prompting ChatGPT to explain its reasoning process can lead to more accurate and insightful responses.

Why is reasoning important? The explanation of reasoning often leads to more accurate results.

Example:

If you’re solving a complex riddle, ask ChatGPT:

“Explain step by step how you arrived at the solution to this riddle.”

This method encourages ChatGPT to break down its reasoning, providing clarity and transparency in its responses.

As you can see, the fact that the model was induced to explain its reasoning to solve this math problem led to more accurate results in chain of thought prompting.

GPT-4 outperforms old ChatGPT in advanced reasoning capabilities, so you may or may not need Chain of Thought Prompting in GPT-4.

Epilogue

Mastering the art of prompt crafting is akin to a wizard perfecting their spellcasting. By employing techniques such as incorporating examples, assigning roles, adding descriptive elements, and guiding through reasoning, you can unlock the full potential of ChatGPT. Embrace these strategies to enhance your interactions and achieve more precise and engaging outcomes.