Let's talk prompt engineering

Welcome to another edition of what we’re determined to make the best damn newsletter in AI. Here we’ll break down AI topics that matter, open your mind to use cases, and keep you ahead of the curve.

Our #1 goal is to be useful. So please shoot us an email 📩 if you have questions or feedback, and especially if you implement something we share!

Here's what we're covering today:

• Applying the scientific method to prompt engineering for better research structure

• How to make Midjourney your next design partner

• Microsoft and ChatGPT developments you should be following

Let’s get to it! 👇

TODAY'S PERSPECTIVE

Apply a prompt research process to your next GPT session

Due to the overwhelming popularity of our Ultimate Prompt Engineering guide and an influx of great questions (btw you can always respond with questions) - we're going to cover how we perform prompt research.

The idea of prompt engineering isn't new - just check out Google's interest over time for "prompt engineering" for the past 12 months.

Investing in discovering the best prompts for your use case is proven to help you get better output. We’ve personally found that it’s best to treat your own prompt discovery like the scientific method.

The Prompt Research Process:

1) Identify what you are trying to achieve and why - explicitly writing out your goal will help guide what tips and methods you need to research in the next step

2) Do background research - this includes looking at prompt guides, like OpenAI's prompt design guide or our Ultimate Prompt Engineering Guide, for tips and inspiration

3) Craft your first prompt hypothesis (or hypotheses plural 😉) - if you're having trouble, use the "If [we use this prompt], then [we'll get this type of output]" approach to writing hypotheses

4) Test it and measure quality of output - quality of output measures will be dependent on your goals, but could include how accurate or persuasive the output is

5) Troubleshoot or come up with new variations - pro tip: ask GPT for variations of your original prompt to help spark new framing or syntax

6) Repeat step 4 - test and measure quality of the new output - prompt iteration is your best friend

7) Analyze all test results - keep a written record of all of your prompts and outputs for side by side comparison and analysis

8) Ship it! - share your prompt research learnings with colleagues or friends and ship it to your end users

USE CASE DEEP DIVE

Inspire your next app design

Historically, platforms like Dribbble, Behance, and even Canva have served as design inspiration for getting started on new projects or giving something you’re working on an aesthetic face lift.

Enter AI.

One of our favorite uses of AI art generation tools like Midjourney is to inspire design. Simply ask the AI to generate a website, mobile app, dashboard, you name it. Experiment with adding terms like "Dribbble", "UI/UX" and "Figma" to the prompt to get better results.

Here’s an example of a generation for a data analytics app (with prompt below).

You'll notice that it doesn’t get the text right - we’ll cover that in a future newsletter 😄

Want to test it out yourself? 

Here’s an entire Twitter thread of designer-worthy Midjourney prompts to get you started.

And here's the Midjourney Guide to Prompting documentation - you're welcome!

LINKS

For your reading list 📚

Microsoft making moves...

• Microsoft releases VALL-E, a text to voice model with ability to fine tune on any voice with only three seconds of training data

• Microsoft looks to add GPT into its suite of products (Word, Outlook, Powerpoint, etc.)

• ... and if that wasn't enough, Microsoft is also looking at investing an additional $10B in OpenAI 

ChatGPT is changing business models everywhere...

• We knew free ChatGPT wouldn't last forever, but looks like OpenAI is thinking the answer is to get SaaS-y; the cofounder opened up waitlist for pro version of ChatGPT Tuesday night

• ChatGPT isn't the only business model in question. The Chaser, a British satirical media company, makes a move to put their content behind a paywall before AI grabs it

And if you're really nerdy ...

• The technical details behind OpenAI's proposed statistical watermark on generated text as unveiled in a talk at the University of Texas at Austin

• A link to a tweet and a paper that showcase X-prompting, a method for prompting a LLM beyond natural language

That's all!

We'll see you again on Tuesday. Thoughts, feedback and questions are much appreciated - respond here or shoot us a note at [email protected].

Cheers,

🪄 The AI Exchange Team