ChatGPT: Practice Makes Perfect
Practice Prompt Engineering and Get More From ChatGPT

Quote 💬

The Best Way to Get Better at Photography? Take More Photos!

bobbyg603

Practice Makes Perfect ☑️

It may sound cliché, but practice really does make perfect. Humans are constantly shaped by feedback, be it the pain we feel from failure, or the satisfaction success garners. Trying and trying again allows us to learn new techniques, build skills, and improve faster than speculation alone. While listening to Atomic Habits by James Clear, a segment that illustrates the value of trial, error, and experimentation piqued my attention.

Photography 📸

A professor at the University of Florida, Jerry Uelsmann, split his photography class into two groups, a quantity group, and a quality group. The quantity group was to be awarded a grade based solely on the number of photos they took throughout the semester, whereas the quality group’s marks would be based on the quality of a single picture. The quantity group spent the semester exploring the craft, tweaking their process, and improving their skills. In contrast, the quality group spent their time either thinking about taking the perfect photo or procrastinating and doing other things. At the end of the semester, the quantity group produced all of the best photos. Learning requires trial, error, and deliberate practice.

Prompt Engineering 🗣️

Large Language Models (LLMs) are a new technology that has engulfed the collective zeitgeist. OpenAI, ChatGPT, DALL-E, and GPT-4 are leading the charge and are good examples of an emerging new skill dubbed “prompt engineering”. According to fourweekmba.com, “Prompt engineering is a natural language processing (NLP) concept that involves discovering inputs that yield desirable or useful results”. As with any new technology, prompt engineering takes a bit of experimentation to figure out which techniques yield interesting results.

Procrastiworking 🙉

This morning, while procrastiworking on a new publication, I decided to open ChatGPT in hopes of getting the creative juices flowing. What transpired was one of the first times I felt I had actually come up with effective prompts, despite fiddling with LLMs for several months. In an effort to reinforce what I’ve learned, I thought it might be interesting to share the series of prompts and the surrounding context.

Analysis 🔬

To start the conversation, I provided ChatGPT with additional context about the result I was trying to achieve. Providing ChatGPT with additional context allows it to answer questions in a manner that’s novel and unique. Additionally, I asked for 3 responses so that I had a few options to choose from.

The results from the initial prompt contained introductory sentences that felt pre-packaged and boring. I asked ChatGPT for a re-do. The result was still not what I was looking for, so I attempted to inject a storyline for it to build off. Injecting a storyline proved to be a good strategy, it added creativity to the response and nudged the result closer to the theme I was looking for.

Finally, the conversation started to converge on something interesting. I picked the second item and asked ChatGPT to expand on its response. After a quick disambiguation, I got a result that I was happy with. Granted, I don’t plan to use this result verbatim, but I’m happy that it was able to get past my writing block and build something that I can reference in my next article.

Takeaways 💭

Here are a few interesting things to try when writing prompts for LLMs:

• Give the AI some background context or ask it to assume an interesting persona.
• Ask the model to answer with more than one permutation, and have it expand on your favorite response.
• Inject a story into the conversation

These techniques are just a few strategies that you can employ to improve your conversations with ChatGPT. If you’re interested in learning more techniques, I’ve found the following email newsletters particularly helpful for keeping up with all things AI. Check out TLDR, The Rundown, and Built with AI.

Thanks for reading!