Behind the Story: Navigating the Art of Prompt Engineering in Generative AI
There is more to a story than the final destination. The journey along the way captures all the small lessons, tangents, and bumps in the road. While writing Navigating the Art of Prompt Engineering in Generative AI I took the obvious approach of using AI, especially as a co-writer, interviewer, editor, and colleague in getting to the point of publishing. Back in early 2023, as GPT 3.5 with ChatGPT was gaining popularity and GPT 4 was right around the corner, I found myself struggling to get what I wanted from GenAI. I could see the magic, but didn’t know how to use it. So I set out to massively change the direction of my career and life.
In early 2023, especially right after GPT 4 was released, I used the idea of AI First as a forcing function for most everything I do. Much to my wife’s annoyance, I turned my default first step in 95+% of my tasks to be consulting with GenAI. It started clunky and an enormous practice in trial and error. Far more error than anything else. But looking back nearly a year now, the opportunities I see and the gains I’ve felt are nearly indescribable. From discovering new terms to me, such as Tacit Knowledge, to introducing others to new terms that push our perception of the future, such as AI Workforce, Generative AI has truly changed my outlook on the future.
In this “Behind the Story”, I’ll walk through the approach I took in collecting my thoughts, narrowing the intended outcome, building examples, drafting, and ultimately posting the article.
An interview with ChatGPT
I’ve found that ChatGPT Voice has dramatically changed how I collect my thoughts. Though I haven’t found the perfect way to translate those conversations to personalized synthetic articles as of yet, that’s a big idea in my head as well.
I’ve found that while ChatGPT cuts me off often while talking, it is a fantastic brainstorming tool to get my ideas out of my head and expanded into the details I haven’t yet expressed. These are not short conversations either, often when I close the dialog, I find that I’ve been in it for 1–2+ hours before I’m cut off from the GPT 4 message cap. 😤
While another post would probably be more appropriate to delve into the details of how I structure such long conversations with ChatGPT, the brief is that I’ll tell it my intended outcome, state the subject matter, and often ask it to ask me questions to expand on these ideas. I also find it useful to ask it its opinion and explore its thoughts to expand my own thinking. This allows me to walk through each topic and question one at a time, exploring other tangents and fully articulating my views.
While collecting my thoughts for the post, I sat with that realization that the conversation indeed felt like an interview, which excited the idea of an AI to conduct one-sided interview style podcasts. While AI may not have the most interesting character development story to lend to a unique interviewer style, that is only a matter of time, so a one-sided will have to do for now. As a result of the conversation, I had a 5,500 word discussion on the topic of Prompt Engineering with my thoughts on the nuances and details I’ve found to be true and useful in developing my lens. As a result, I asked GPT to tell me how many topics were large enough from that discussion to be their own article, it told me twelve topics, nine of which resonated with me.
Co-writing Discussion
Now that I had a general conversation about Prompt Engineering and a lot of the details I care about, the next step was to narrow the scope and focus on one area to cultivate. Similar to the initial brainstorming session, this time I gave GPT the contents of the first discussion and told it the narrow topics and to write an outline that we could discuss to begin writing the draft of the article. From this second level discussion, I came away with another 8,400 words about writing the article itself and the details presented therein. This allowed me to keep in my more natural conversational flow while curating the content further for the intent of the article. While the words may not be exactly how I’d normally phrase things, the ideas, mental models, and thoughts are my own. I’ll get to this a little later as to how I see the blend of Human/AI writing and bias behind it.
Building Examples
While I’m writing the article, there were a number of examples I wanted to convey to fully capture the meaning. But I had a hard time thinking of actual scenarios when talking with AI. And so I fell back on my typical approach now, which is to ask AI to give me ideas for such examples. And it worked out great! It gave me the two examples, and I was able to use my approach to solve those specific needs.
The example of the marketing slogan helped me to highlight the difference of how much information is given and the output from each open AI model. And while I prompted it to give me the questions to narrow down and clarify the intent of the company for the slogan, I also told it to make assumptions and answer those questions itself. This is another example of the meta feedback loop that I find within AI to expand on its own thoughts and speed up my output significantly.
The second example was equally fascinating because after doing the first example, I ended up using the slogan as the input for the second scenario. The problem was that I didn’t know how to categorize the term slogan, or even what other categories would be close to it. To solve this, I asked GPT to give me other words like Slogan and it didn’t do so great. Then I listed the words “Title, Description, Tagline, Slogan” and asked it to give me more examples of similar words. This gave me exactly what I needed, a list of 40 terms that are similar in categorization with slogan that I could use as the data source.
The best result was the moment I put the actual slogan in and ran the text analysis prompt, the first category it returned was in fact slogan. It was a nice moment of convergence between these two examples to show increasing complexity, but also the usefulness of AI in this narrow scope.
Editing
My initial approach to editing is straightforward, largely because I have not yet developed a detailed set of standards for the process. Currently, my method involves simply instructing GPT to examine the text and suggest any edits it deems appropriate. Given GPT’s tendency to be excessively positive — often to a fault — while its affirmations may boost self-image, my objective remains focused on identifying and implementing necessary revisions.
The outcomes of this preliminary process have been promising, encompassing a range of improvements from correcting spelling and grammatical errors to enhancing sentence structure. GPT has also proven adept at suggesting organizational refinements, such as arranging content into numbered lists. These outcomes suggest that, over time, I will be able to craft a more comprehensive set of editing guidelines tailored to uphold a standard of precision and attention to detail that resonates with my personal values.
My ideas vs my words
I’ll do a little experiment here. I find I can expres myself and my thoughts more effectively over voice. So to complete this section I used Speech To Text to capture the below and will use AI to rewrite it into what I feel is a more eloquent expression of my thoughts.
I think that the topic of AI vs human writing is quite fascinating and reminds me of the arguments when people started to write with implements less and use keyboards more, such that the originality and skill is only in one place. Some people have amazing penmanship and artistry to things like calligraphy. And it may be that they are writing words that originated from other people, but that those words are more impactful when captured in that style, that visual style of writing. Whereas someone might type words using something like a voice-to-text tool, because they are unable to articulate a writing implement, still capturing the masterpiece in their mind. It feels to me like AI is the next step in this, where some have developed an amazing appreciation for vocabulary and the sequence of words to convey a meaning with poetic nuance that said any other way would not capture their brilliance. With the augmentation of AI, it feels to me like an opportunity for some to focus on the meaning and ideas they want to convey without worrying about or having to focus on the exact words, but rather delving into the depths of knowledge to find the most impactful and highest density knowledge that they can.
Obviously that was a long string of words without structure in mind, just a stream of my thoughts. Now here is the same content using AI to give it more structure, expanded expression, but keeping with the meaning I intended.
The topic of AI vs human writing has always fascinated me, evoking memories of the debates that arose when people transitioned from hand-written texts to keyboard typing. It was a time when the notion of originality and skill seemed to reside solely in the act of physical writing. Some individuals possessed exceptional penmanship and artistic flair, particularly in disciplines like calligraphy. These individuals might be transcribing words that originated from someone else, yet their unique visual interpretation breathed new life into those very words, imbuing them with a greater impact.
On the other hand, we have those who rely on voice-to-text tools to compose their thoughts due to various constraints. Though they are unable to manipulate a writing instrument, they still manage to capture the essence of their creative brilliance through typing. In this scenario, the medium of expression becomes secondary as they tap into the depths of their mind to generate a masterpiece.
In addition to the aforementioned groups, there are two more distinct categories worth mentioning. Firstly, there are those who possess an innate talent for ingeniously stringing words together, akin to skilled artists. For them, AI serves as a powerful tool that aids in expanding their vocabulary, refining their flow, and enhancing their linguistic finesse. With AI’s assistance, their written expressions radiate even more brilliance, painting vivid pictures in the minds of their readers.
Secondly, there are individuals who prioritize delving into the vast depths of knowledge, seeking to explore new words, ideas, and concepts. AI becomes their ally in this quest, offering an unmatched capacity to sift through an immense sea of information. It empowers them to unearth the most fascinating and thought-provoking nuggets of wisdom, expanding their intellectual horizons and providing them with fresh perspectives to weave into their creations.
By supporting all of these groups, AI enables writers to pursue their unique passions with even greater intensity. Whether it be mastering the art of linguistic arrangement, etching ink to parchment with artistic flourish, voicing inner musings, or unearthing hidden intellectual gems, AI’s contribution is invaluable. It serves as a facilitator, empowering each writer to reach new heights of creativity and fulfillment in their chosen path. Through this harmonious partnership, human expression not only transcends the limitations of conventional writing but also embarks on an awe-inspiring journey towards unprecedented levels of innovation and artistic beauty.
But why…
I wrote this Behind the Story post as I felt there was more to say about how AI can be used and understood. My hope is to help you learn from the trial and error I’ve gone through and perhaps give you some clever approaches and tools you can use to leverage AI to a greater degree. I see a world changing wave of change within the Age of AI, and as I’ve benefit from others, I wish to give back to help others along this change so we all live more fruitful lives in the future.
