To some, ChatGPT can seem like an omniscient oracle. You ask it a question and it spits out an intelligent response in its AI-speak. And just like that, you are done: in hand, all polished and shiny you have a response.
But working with ChatGPT isn’t about pressing some magic button just once to get what you want. It’s actually about “interacting” with the AI. After all, the “Chat” in GPT came about because the creators–OpenAI–weren’t just expecting its users to have a conversation of sorts; they knew that the back and forth between user and ChatGPT would give it the most input to work with, in essence refining its understanding of what the user wanted.
However, even those who know the technology well, and tend not to think of it as some creepy deity, are often inclined to ask it only several questions—or what are known as “inputs” —rather than to “chat” with it.
In this post, I’ll teach you how to transform your basic exchanges into rich, outcome-driven conversations, or what I call a “prompt-versation.” The net effect? A more polished final output, a deeper understanding of what you are learning, and a new perspective on a complex issue.
Creating Prompt-versations
So, what does an actual prompt-versation look like?
I’m going to provide three examples, each followed by a verdict. But before you read the verdict, you should judge yourself how strong you think each thread is in terms of the quality of outputs given. I’ll rate each on a scale from 1–10, and you can do the same if you like!
Example Thread 1
Verdict:
This is not a terrible thread. It gets some of the basic information out there, and in the series of exchanges is able to elicit a basic gameplan. But it’s a little too casual and not the kind of conversation that’ll get you the useful outputs you are looking for.
By conversation, I mean something a little bit more professional but still chatty, perhaps the way you would speak to an actual educator or expert. In this case, the conversation could have started much more effectively by dispensing with the unnecessary introductions and spelling out the entire issue in one prompt, as we see below.
I’m new to AI and feel overwhelmed by all the acronyms. It’s difficult for me to know where to start. What would you recommend?
Score: 6/10
Example Thread 2
Verdict
This is by no means a bad output. It is solid and offers a good starting point. But how targeted is it? It is after all, what is known as a “one-shot” prompt: there’s one input and one output. What if the user isn’t as focused on the Ethical Considerations? And what exactly is their goal? This is something that might be fleshed out in a “prompt-versation.”
One way to improve the prompt is by providing a follow-up prompt to help the AI provide more targeted advice. There’s also some more subtle things that we’ll touch on in thread #3 below.
Score: 7.5/10
Let’s now compare this to the third and final thread (again, I’ve shortened some of the snippets for readability).
Example Thread 3
I’m going to jump in mid-thread to make a few quick comments! Notice how the very first prompt starts off by setting up an exchange that is going to go on for more than “one-shot” (one exchange): “…before you do so, can you let me know if there’s any other information I should provide…”
This allows the user to refine their initial prompt to allow for a much more targeted and refined output. Let’s take a look!
I’m not going to reproduce the entire curriculum here, but you can see just from “Month 1” that you get a bi-weekly by bi-weekly breakdown of what to do. This is super helpful and structured. And it is because we had prompted ChatGPT to help us refine our prompt, that it coaxed out the important stuff — commitment level, learning style — that is oftentimes easy to forget when we create an initial prompt.
One interesting thing: even though, we had asked the AI in thread #2 to “act as an AI educator”, it’s here in the prompt-versation that this happens naturally.
And, there’s actually one REALLY interesting thing. Notice the level of the language in thread 3 is more conversational than thread 2. The responses from ChatGPT themselves are more conversational (notice the use of the exclamation marks). This human touch, I believe, makes it much easy–and more natural–to have a prompt-versation. You are essentially conversing, not writing out a list of commands in stilted prose and getting that reflected back at you.
And that dynamic makes it much easier to have a back and forth, in which relevant information is more likely to suddenly occur to the user. For example, let’s take a look at the rest of thread 3 (yep, we’re not quite done!)
Now, from this natural back and forth, the user has a customized curriculum and a roadmap (in the partially reproduced snippet, right above) on how to incorporate a social angle into their learning journey.
Verdict
This thread, spread out across several outputs, is much stronger than the first two. It offers targeted, specific advice and even a customized curriculum. Sure, there could be a higher level of specificity and something even more targeted around understanding enough AI to lead a team, but this is exactly something the user can follow up on. The prompt-versation doesn’t have to end there.
Score: 9/10
What did ChatGPT think?
One final thing I did, just to benchmark my own subjectivity, is feed these three threads into a new ChatGPT thread. Here’s the AI’s quick synopsis on the three threads:
Takeaways
Getting the best outputs comes not from one exchange, but a back-and-forth dialogue between user and the AI
Using a conversational, but professional tone allows “prompt-versations” to happen more naturally, as the AI will often reflect this tone back at you
If an output isn’t quite specific enough for you, always feel free to follow-up with more details
Action Items
Try to aim for at least three follow-up inputs (after your initial input) that relate back to the AI’s responses.
For follow-up prompts interact and respond to the AI’s outputs; this can often propel you in new directions, leading to inputs you otherwise might not have had.
Lastly, curiosity is often your best guide in “prompt-versations”, letting serendipitous discoveries emerge.
Comments