Tell a simple story to an AI like GPT-3, and it turns into a personal coach.
A writer who listens to feedback and makes changes to a draft.
‍An art director who offers advice on styling your images.
‍A partner who can stimulate and challenge your thinking.

Anything you want, essentially.

To do this, you don't need technical skills, only the ability to write English, a free OpenAI account, and some inspiration.

In this two-part series, I will provide examples, templates, and tips on how to talk to an AI to make you more productive and creative.

Template #1: The Expert Generator

The first template summons an expert on a particular subject, whom you can then ask one or more questions.

Before we dive in, here's how to read these templates:

• Yellow sections are placeholders for your input (in this template, the question we want to ask the expert).

• Black text on a white background also goes into GPT-3. (See the end of this email for instructions and a video on how to do this.)

• Orange blocks with "🤖 Start AI" indicate where to start GPT-3 to generate its output.

• Green highlighted parts are where the AI's output goes.

Now, let’s look at the Expert Generator in action. The image below shows an example—I have a question about making prompts for AI image generators.

To recap what’s happening above: First, we use the AI to determine the qualifications needed to answer the question. Then, we ask the AI who has such expertise. Finally, we ask that expert our question.

This first example offers several insights into what you can accomplish by talking to an AI:

• Create a "machine" that follows specific rules by telling the AI a story. In this case, the Expert Generator that "will find the person most qualified to answer the question."

• Let the AI help you identify the required characteristics, conditions, or qualities something should have, even in an area you know nothing about. In this example, the experience an expert in AI image prompt creation should have to answer my question.

• Get better results by steering the AI in a specific direction. In this instance, we direct it to answer from the perspective of one expert instead of its entire body of data.

To understand the importance of this last point, we need more background on how GPT-3 works.

When you give GPT-3 an assignment, it looks through all the data it's been trained on (45 terabytes of text, the equivalent of ~5.6 million books or ~10 million Wikipedia articles). It searches for patterns in that information to produce the output it believes is most likely correct or desired.

Depending on your question, the AI might take a kind of average of different possible answers. That approach can give a satisfying result, but the outcome can also be bland (or even false).

Asking GPT-3 to simulate an individual or a particular method can ensure a focused answer from that unique perspective.

Template #2: Improve a draft email or article based on feedback

In this second example, we have feedback from someone on a piece of text (like a draft article or email newsletter) and let GPT-3:

1. Summarize the action points from that feedback.

2. Rewrite the text for us based on those action points.

Here’s the template:

See this template in action here. Scroll down to the Example section to see a draft I wrote, the extensive comments someone sent me, and how GPT-3 does a great job summarizing that feedback and rewriting the newsletter.

Takeaways from this second template:

• Feed GPT-3 long, unorganized pieces of text, and it can still summarize and extract action points.

• Break your request into smaller steps for better results. In this example, we first ask the AI to find the action points and then use a second prompt to rewrite the text instead of squeezing all that into one request.

• Set rules for how the AI should perform tasks and understand information simply by talking to it in English. In this example by explaining how to recognize the newsletter's start and section headings.

Five fundamentals to remember from these examples

The five fundamentals to remember from these two templates and their examples:

1. Tell the AI a story to create a machine that does what you want.

2. Let the AI help bridge knowledge gaps by asking it qualifying questions about a topic or skill.

3. Send the AI in a specific direction to prevent bland output.

4. Create rules by explaining them to the AI in plain English.

5. Break requests into multiple steps instead of trying with one perfect prompt.

How to use GPT-3 and these templates? 🤷‍♀️

To use these templates or make your own, go to OpenAI's GPT-3 Playground and follow these steps:

1. Create a free OpenAI account.

2. Copy the text from the top of the template to the line above the first orange box that says "🤖 Start AI," and paste it into the Playground's input window.

3. Complete any sections marked in yellow that require your input.

4. Place your cursor after the last text character without leaving a space.

5. Press the Shift + Enter keys on your keyboard, or click the Submit button.

6. Wait for GPT-3 to produce its output.

7. Delete any output from GPT-3 that’s unnecessary or irrelevant.

8. If applicable, copy and paste the second part of the template into the Playground input window. Start after the first placeholder for AI output and stop before the second orange box.

9. Fill in the sections marked in yellow that need your input.

10. Repeat from step 4 until there are no orange boxes with AI starts left.

Click the image below to watch a video where I explain these steps using the Expert Generator template as an example. 👇️

I hope you're now inspired to try out these templates and create your own AI machines.

Please share with me what you create, and send me comments or ideas for improving these and other templates. Just reach out on Twitter.

In a future weekly email, I’ll share part 2 with more advanced ways of talking to an AI and getting the results you want.

‍