Prompt Engineering or AI Wrangling? How to Work with Your New Tool

/

Every good blog post needs a hook. For today, I wanted to leverage the continued growth of ChatGPT registered users. Given it hit 1 million in 5 days, where must it be now? Likely some big, juicy number that would be great for this post. I asked Chatsonic “where can I verify the number of registered users chatgpt has today?”, but it told me yet another in a long line of fibs:

“You can verify the number of registered users on ChatGPT's website. On the homepage, you can find the "Users" section, which shows the total number of registered users on the platform. You can also find the total number of active users in this section.” […there is no Users section on either the OpenAI or ChatGPT sites.]

Is it Me?

Maybe I just don’t know how to ask the right question… but I’m likely better than many. I’m known in the family as the Queen of Search (thank you, Google Advanced Search course!) and much of my research involves me crafting search requests across a variety of platforms. It may also help that searches early in my schooling cost actual money and I had to make an appointment with the university librarian to submit my request (1980s Dialog searches for academic articles).

We’re All (including the AIs) Learning as We Go

The requests I’m running these days cross ChatGPT, ChatSonic, MidJourney, DALL-E 2, and more. Each of these platforms has a slightly different set of parameters you can use. How about this one for MidJourney: “a seamless tileable jade tree pattern, spiral carvings, octane renderer, trending on CGsociety --ar 1:1 --q 2”. Not a phrase I would have come up with, but the result is lovely. There is also the complexity of integrating some of these capabilities into our standard tools (e.g., ChatSonic extension for Chrome).

Thank you to those who have gone before and their willingness to write instruction guides -- see my short list below. Thank you too, to the platforms for offering up backgrounders to help us learn the nuances and see more opportunities. I’ll list some of those below as well.

What is the Expertise We Need?

Queen of Search may be a great title, but I wonder if we will find it on many LinkedIn or Indeed job posts. Prompt Engineer is an actual job title, but I’d like to see it fade away. Two reasons: First, I’d like all my engineers to be prompt, and second, given we are unlikely ever fully to understand the AI black boxes, is it engineering? 

How About AI Wrangler?

AI Wrangler builds on the short history of Robot Wranglers. It also reminds me of the skills needed to herd literal or figurative cats. Until, if ever, we know what it means to be fluent in the language of AI, wrangling is more in line with the reality of this work than engineering. 

Learning from Others

Prof. Ethan Mollick offers examples that show me that I have much to learn as I work with these new tools:

write an essay with the following points. use an academic tone. use at least one clear example. make it concise. write for a well-informed audience. use a style like the New Yorker. make it at least 7 paragraphs. vary the language in each one. end with an ominous note. -Humans are prone to error -Most errors are not that important -In complex systems, some errors are catastrophic -Catastrophes cannot be avoided….

Sometimes the memory is useful, you can (and should) ask it to revise previous work: change the third paragraph to be more professional or use a different example in the middle and it will provide a revised version. However, sometimes it forgets what you were working on, and you have to remind it. You may, for example, want to tell it revise the third paragraph on the essay on catastrophes so it doesn’t lose track.

And It’s Not Just the AI - We Need to Work Collaboratively

Screenshot of a ChatGOT answer with the thumbs up and down icons

ChatGPT response with opportunity to provide feedback via the icons

Yet, focusing on prompts seems limiting. If my goal is to race with the machine, I want to see the process become more of a collaboration. Alexa and Google Home sometimes offer options after answering questions, but usually these are more of an annoyance than a help. One more thing to say “no” or “stop” to. Coaching from the new AI would help me. Live-action versions of the backgrounders would be nice. Humans learn best actively -- when we’re in the middle of a task is the best teachable moment. ChatGPT seems the closest of the tools I’ve used in that it asks for feedback on the quality of the response. Sometimes, it also gives a hint. In this screenshot, it coaches me on how to find what I’m looking for.

My Attempt to Be Proactive

I asked ChatGPT and ChatSonic to give me hashtags for my last post. “Act as a coach, how can I more effectively ask you for help in creating good hashtags for this post?”

ChatGPT’s response:

1. Be specific: Provide as much context as possible about the post, including the topic, audience, and goals.

2. Provide an idea of what you're looking for: Are you looking for hashtags that are popular, unique, or specific to a certain community?

3. Be open to feedback: Be prepared to receive suggestions and be willing to make revisions based on my feedback.

4. Give a good example of the post, if possible. Tell me if you have any specific guidelines that you need to follow, such as the number of hashtags, length, etc.

5. And finally, be clear and concise in your question. For example, "Can you help me create 5 hashtags for a post about vegan food targeting a health-conscious audience?"

Learning Together

We are writing thousands of articles in hundreds of languages about using these new tools. (WriteSonic even offers example prompts for AI-related posts in its menu bar.) I’m reading and experimenting with as much of this as I can. I hope my colleagues and students are as well. Every time I write, stumble at getting the AI to do what I want, or comment on a colleague’s related post, I’m hoping this work is training the various AI as well as giving me a chance to practice. I’m willing to put the work in now for better tools in the future. 

Take your AI wrangling on as a challenge. If you spend much time interacting with voice assistants like Alexa and Siri, or maybe your smart oven, you know that (at least today) you must adjust your language. You may adjust your pace, tone, and vocabulary to get the desired result. Working with the new batch of generative AI is the same. …and these are the same skills you may have had to develop if you’ve programmed in anything from COBOL to Python: We start with a “Hello, World” program and move on from there (…and have no worries, ChatGPT can write those “Hello, World” programs for you and explain the work -- just ask).

The good news is that, like with programming in different computer languages, once you learn one, the next is easier. Consider other life skills that you can draw on as you learn these AI languages. In a constructive Medium post, Shubham Saboo says, “In a way, prompt design is like playing a game of charades!” Unfortunately, if I’ve played charades, it’s less than three times. Good thing I have COBOL to fall back on.

Thank you for letting me share my learning journey.

Resources

Each of the following is a support document I’ve found helpful. Thank you for sharing others you find (or that you had the AI find for you) in the comments section below.

Embed Block
Add an embed URL or code. Learn more