In the last three posts (1, 2, 3) we evaluated ChatGPT’s ability to create compelling text for a book about retirement income.
Most recently I felt the need to invent two characters whose financial situations and household conversations are intended to include all the boring factual details that ChatGPT was able to provide, but with the addition of a recognizable human spark that ChatGPT seems to lack.
Our fictional couple, Carla and Matt, ultimately became the framing device that helped me humanize ChatGPT’s text. Using character-based prompts, I generated AI text for the next sixteen chapters (the original outline called for 18 chapters and an afterword) using Carla and Matt as our retirement income education avatars, and met with similarly moderate levels of success as chapters 1 and 2 as seen in the previous posts.
None of the text that was generated was worthy of being published. But increasingly I began to see how ChatGPT could be a functioning writing assistant rather than a writer.
Prompting is as annoying as working with any other form of software — you must adapt your brain to its limitations and constraints to gain from its technological power. Just as I have learned how to write Excel formulas, so I pushed forward to learn how to squeeze usable human-centric text from my nonhuman companion.
Automating text is like any other form of automation: it must be supervised. After repeating the above processes and ultimately generating 18 chapters of material, I began to edit and rewrite and re-prompt as necessary, leading to several conclusions:
ChatGPT is not creative — it relies on boilerplate and cliché to communicate.
Facts are essential but secondary; in a nonfiction personal finance book the context of why these facts should be important to me is primary. Functional accuracy is simply not engaging.
It’s not the heat, it’s the humanity — humor, empathy, and instinctively recognizable human situations will upstage even the warmth of accurate information every time.
Send in the clowns
To gain the most from ChatGPT, it appeared that I would need an ensemble cast — not just Carla and Matt. To illustrate financial situations beyond a pre-retirement married couple the book would require friends, neighbors, and ultimately a financial advisor to join the various conversation that would comprise the bulk of the book. None of these character ideas came from ChatGPT but rather from me, a non-AI, as I began to wear a playwright’s hat instead of an author’s: inventing characters with entrances and exits, and the situations and conflicts they experience.
Into this larger ensemble context, it began to become apparent that ChatGPT was capable of threading the material into these new characters’ lines like beads on a string — producing personal financial insights in the context of the conversations of our protagonists as well as their friends and neighbors. Less blandness and less matter-of-factness opened the door to more jokes and maybe even something like poignancy.
Except that ChatGPT was failing to deliver jokes or poignancy.
Less cringe, please
At a certain point I decided that Carla and Matt needed to have a cat. (Feline interest in retirement in implied by the book’s title: I Don’t Like Cat Food: Making Your Money in Retirement Last As Long As You Do.) Here is my conversation with ChatGPT on this new character (prompts in blue, ChatGPT in black text):
Carla and Matt have a cat, who is surprisingly wise about financial matters. What is the cat’s name and backstory?
Let’s create a whimsical backstory for Carla and Matt’s wise feline companion:
Name: Professor Whiskers
Backstory: Professor Whiskers isn’t your ordinary cat. Found as a stray kitten wandering the streets, Carla and Matt took him in, never expecting that he would become their unexpected financial advisor. With a distinguished gray coat and piercing green eyes that seem to hold a lifetime of wisdom, Professor Whiskers earned his name through his keen insights and sage advice on matters of finance.
No one knows where Professor Whiskers came from or how he acquired his remarkable financial acumen. Some say he was once a Wall Street trader who retired to a life of leisure as a housecat, while others speculate that he’s a reincarnated financial guru with a mission to guide Carla and Matt towards financial success.
Regardless of his mysterious origins, Professor Whiskers has become an integral part of Carla and Matt’s household, offering guidance on everything from budgeting and saving to investment strategies and retirement planning. With a flick of his tail and a purr of approval, Professor Whiskers helps steer Carla and Matt towards a prosperous future, proving that wisdom truly does come in all shapes and sizes, even in the form of a feline friend.
I like the cat’s backstory but not his name. Please give him a one-word name that would not be inappropriate for a human but is offbeat and unusual.
Name: Eccentric
That’s pretty good but give me 6 to 10 more options.
Certainly! Here are some more offbeat and unusual one-word names for Carla and Matt’s wise feline companion:
1. Whimsy
2. Nimbus
3. Seraph
4. Quirk
5. Zenith
6. Mystique
7. Apollo
8. Solstice
9. Echo
10. Zephyr
•
Let’s parse these responses for a moment. ChatGPT did not come up with the creative idea of a financially astute cat. When I asked for a name suggestion, I received “Professor Whiskers,” which might be ideal for a toddler’s board book or an Andrew Lloyd Webber musical, but perhaps not for a personal finance book. When I asked for a one-word name, its suggestion was “Eccentric,” which it certainly is. When I asked for ten more suggestions, I finally found something that was usable, or at least in accordance with my taste: Apollo.
Apollo is a name I would never have thought of myself. Its reference to mythology matched the mythic role a cat is supposed to play in a personal finance book if it appears at all, which is exactly what I was hoping for.
The cat’s name, Apollo, stuck.
Later I invented the remainder of the cast of characters, and ChatGPT helped me develop backstories that were suitable. Could I have done this myself? Of course. But, like the name Apollo, ChatGPT gave me certain options that I would never have independently arrived at, given my history and my conceptual habits.
This is what I was looking for: a new way to refresh my own imagination, and (possibly) augment my production of text.
Fast ≠ forward
I had reached the point where ChatGPT took my original outline (x,000 words) and, after much wrangling and false starts and numerous creative leaps on my part, managed to produce its own text of xx,000 words.
I suppose it was better than the nothing I started with, and possibly improving, but nowhere near ready for human consumption. It reminded me of those “Do Not Eat” silica gel packets that come with consumer products — a touch of desiccation for your own good.
The text had reached a wall instead of a crossroad. Further tinkering would only bring more of the same: unimaginative but largely accurate implementations of the ideas I was feeding ChatGPT.
Was it worth the effort? Maybe.
As I said, the AI occasionally came up with small but relevant ideas that I felt advanced the text. But was the text worthy of a book? No.
I began to regard ChatGPT as the other half of my writing group with only two members: somebody off whom to bounce my ideas. Part writing assistant, part cheerful friend, part contributor — all with no ego or agenda or attention span issues.
Pretty cool, right? Yes. But still the book I wanted to see never got written, at least thanks to my capacity as a newbie prompter.
What it’s gonna be… Is AI a fantastic tool or an anxiety-producing authorial substitute? A useful new writing method, or a time-wasting cul-de-sac?
What could I do now to complete the book and resolve my AI issues?
Curious? Contact me. Find me on LinkedIn.
Coming next week: Step 5: Human insurrection
You may find of interest:
Step 3: Character references [6th in this series]
Step 2: Injecting personality into AI text [5th in this series]
Step 1: Giving the outline to AI [4th in this series]
I asked AI to write the book I never got around to [3rd in this series]
Ways to use generative AI text in marketing [2nd in this series]
Beyond perfect [1st in this series]