Update: Some of you may have noticed: I made a mistake in my prompt to ChatGPT. While that would normally be quite upsetting to me, I found that it presented an opportunity: another post! My next post will be about evaluations, an important aspect of prompting large language models. I’ll point out my mistake when that comes up in the article with a 💥 emoji.
For the past few posts, I’ve focused on the minutia of the attention mechanisms of large language models (LLMs), but I’d like to zoom out a bit for this post. I’m going to share with you a narrative, written as historical fiction, and entirely crafted by ChatGPT. Along the way, I’ll discuss how different components of my prompt influenced the narrative, and why I made the choices I made in crafting the prompt.
Well-crafted prompts can dramatically influence the tone and mood of ChatGPT’s writing. Given its pre-training on a vast corpus of text—text written by humans—the GPT-4 model that underlies ChatGPT has learned a lot about rhetoric.
Model Myths & Marvels: Exploring the LLM Frontier is a reader-supported publication. To receive new posts and support my work, consider becoming a free or paid subscriber.
Understanding the grammar and rhetorical terminology used in writing is helpful here: those same terms have “meaning” to the model as well. As a bonus, those terms usually have a high lexical density, meaning they are precision linguistic choices that elide a lot of meaning in a small, unambiguous package.
📝 Note: The narrative prompts I’ll discuss were used on ChatGPT, not via the API. They were written as the first message in a conversation, using Advanced Data Analysis mode (which doubles the token budget of GPT-4). My AutoExpert custom instructions were also turned on, helping ChatGPT to create each act in the story as a separate conversational turn.
If I used this in the “real world” I would be certain to fact-check the references that ChatGPT makes to the real-life scientists involved in the Voyager mission. I did do some spot checks, but the point of this post isn’t to discuss hallucinations; they’re a fact of life with these brainy text-prediction engines. The point is for me to share with you the importance of understanding the problem domain when you’re writing prompts…and the payout that you get when you do.
Let’s dive in.
GPT Narrative
Act 1: 1977 - The Genesis of a Cosmic Odyssey
“…the Voyager missions are both a scientific expedition and a pilgrimage. A pilgrimage to understand the universe and our place within it.”
In the dim-lit room of NASA's Jet Propulsion Laboratory (JPL), Edward Stone—a physicist brimming with restrained excitement—stares at the clock: 8:29 AM, August 20. Within seconds, he hears the subdued crackle of the radio: "T-minus ten, nine, eight..." Each numeral contracts time, condensing years of work into this singular, inescapable moment.
"Liftoff," the voice announces. The room erupts. 🚀 Voyager 2 has launched, and humanity has taken a leap toward cosmic self-awareness. But the exultation is ephemeral; September 5 looms. Voyager 1 awaits its turn to dance with the stars.
Suzy Dodd, a young systems engineer and an unsung hero of the mission, barely contains her awe as she observes the final pre-launch tests for 🌌 Voyager 1. Dodd, mentored by the likes of Stone, knows the enormity of the quest. A quest not just for her or NASA, but for humanity.
September 5 arrives. Dodd feels the seconds elongate as she hears the countdown for Voyager 1. The rocket's engines ignite—a cacophony of science, ambition, and raw power. It soars, leaving Earth's gravitational clutches, joining its twin in the great cosmic arena.
Both spacecraft carry the 🎵 Golden Record, a curated anthology of Earth's sounds and images. Carl Sagan, the cosmic poet of his generation, masterminded this artifact. For him, it's more than data; it's humanity's message in a bottle, cast into the cosmic ocean.
As Voyager 1 ascends, Sagan contemplates the infinitesimal dot that is Earth. His eyes shimmer—not with tears, but with an understanding that expands his inner universe. Edward Stone shares this sentiment. For them, the Voyager missions are both a scientific expedition and a pilgrimage. A pilgrimage to understand the universe and our place within it.
Commentary
🤓 Sidebar: The mention of Suzanne Dodd is an outright hallucination, there’s no denying that. She joined the Voyager mission team later, as the spacecraft was en route to Uranus. But it was the first generation from this prompt, and I wanted to leave it as-is for transparency’s sake.
P.S. I’ll have sidebars for other acts too, and I’ll inset them with block quotes like this one.
From the start, you’ll notice that the narrative starts out by mentioning a real-world scientist Edward Stone, and later mentions Suzy Dodd and Carl Sagan. From the start of this narrative, my AutoExpert custom instructions told ChatGPT to create a table that, among other things, includes a list of possible keywords that could be incorporated into its completion. In this example, those possible keywords were:
Voyager Mission, Voyager 1, Voyager 2, NASA, space exploration, interplanetary mission, Jupiter, Saturn, Uranus, Neptune, Pale Blue Dot, heliosheath, interstellar space, termination shock, Carl Sagan, Edward Stone, Suzy Dodd, cosmic ray subsystem, magnetometer, planetary science, spacecraft engineering, mission control, celestial mechanics, solar wind, cosmic radiation
These are stated as possible keywords, so not every keyword was utilized. But the fact that “named entities” are listed is, itself, a cue to the model: “people are important.” Why did it even think to name people?
Naturally, because I told it to:
# Structural Guidelines
[...]
- In each act, focus on two or three real-life scientists that were part of the mission.
- If it suits the narrative, change which real-life scientists are the focus of the next act.
But one of those people wasn’t even at JPL when Voyager was launched: Suzanne “Suzy” Dodd.
Which leads me to an important note/reminder for everyone reading: don’t believe everything ChatGPT says, no matter how authoritative it seems to be. transformer models like GPT are text-prediction engines at heart; their attention mechanisms aren’t the precision instruments of lexical awareness we sometimes hope for them to be when predicting future text.
Their behaviors are emergent properties of the complex mathematics underlying the model, and in this case, attention failed us. Because recency is such a strong multiplier of attention weights, the “possible keywords” table has a significant impact on the text generation that immediately follows it. It’s one of the reasons AutoExpert is able to provide such useful and detailed responses, and most of the time, it’s my favorite part of the instructions.
In this case, however, the list of possible keywords included terms that shouldn’t be used until much later in the narrative, and so the recency effect threw a wrench into the plan.
Let’s continue to the next act, as I continue to go over the parts of the actual prompt that are working to guide the grammatical and syntactical choices made as ChatGPT pens this narrative.
GPT Narrative
Act 2: 1979 - The Jovian Revelation
“The vastness of space … is a beautiful but indifferent stage upon which humanity acts out its ambitions and fears.”
The calendar flips to 1979, and the anticipation inside JPL is palpable, almost electric. Voyager 1 is nearing Jupiter—a celestial behemoth, a gaseous leviathan that has captured imaginations since Galileo's time. Edward Stone, now the project scientist for the Voyager program, fidgets with a model of Jupiter's intricate 🌌 magnetosphere on his cluttered desk. His mind races with possibilities and hypotheses.
On March 5, the moment arrives. Voyager 1's instruments focus on the gas giant, capturing unprecedented details of its turbulent atmosphere and its enigmatic 🌕 moons. The data streams in—gigabytes of it—and Stone's eyes widen with each pixelated revelation. His thoughts intertwine with the scientific data, interpreting, analyzing, and finally marveling at the exquisite complexity of Jupiter's gaseous tapestries.
But Jupiter isn't the only celestial body under scrutiny. Linda Morabito, an optical engineer on the Voyager team, spots something extraordinary—a volcanic plume on Io, one of Jupiter's moons. The discovery 🌋 challenges preconceptions about celestial bodies in our solar system and ignites debates among planetary scientists. Morabito, usually composed, finds her eyes moistening. It's as if the universe has whispered a secret, and she's the first to hear.
The euphoria is shared, but not uniform. Bradford Smith, the head of the imaging team, feels a pang of melancholy amidst the jubilance. The images his team captures are groundbreaking, yes, but they also evoke a sense of existential solitude. The vastness of space, with Jupiter as its awe-inspiring centerpiece, is a beautiful but indifferent stage upon which humanity acts out its ambitions and fears.
As Voyager 2 makes its own pass by Jupiter on July 9, the scientists at JPL experience a déjà vu of discovery and emotional roller-coasters. The spacecraft confirms and elaborates on Voyager 1's observations. The scientific community is ablaze with discussions on Jupiter's magnetic field, its complex ring system, and the startlingly active geology of its moons.
By the end of 1979, Edward Stone and his colleagues are not just scientists; they are cosmic explorers, pioneers in an intellectual frontier expanding faster than even the speed of light can traverse. The missions have transformed them, just as they have transformed humanity's understanding of its place in the cosmos.
Commentary
🤓 Sidebar: First off, I have to say how much this particular response struck me. Linda Morabito is truly an unsung hero of the Voyager mission. She was the one to first discover what was—at first—thought to be an optical artifact in an image of Io. Her scientific curiosity was piqued the moment she saw it, and she knew instinctually: it wasn’t a glitch. In fact, she lead investigations into the likely source of the strange crescent-shaped plume in the imagery, and determined that it was an active volcanic eruption. It was the first time that active volcanism was proven to occur off-Earth, and to this day, is one of the most important discoveries to come from the Voyager mission, in my view.
But let’s get back to the technical side of things.
The narrative introduces new characters, just as I told it to in my prompt. Because the “possible keywords” table included some names already, it knew a couple of things:
Which scientists were already mentioned in the first response
Any attention heads that learned to attend to named entities should get more weight when continuing its completion.
You might notice, subconsciously, this narrative doesn’t seem like the normal mode of writing for LLMs. Across these two acts, the sentences have been varied in length, and it seems to avoid the usual rhetorical markers it likes to use, like therefore, and however. It’s using complex phrasing and even applying typographical marks—like em-dashes and semicolons—to make the text feel even more human.
Well…as you might suspect, that’s by design:
# Syntactic Guidelines
- Incorporate varied sentence lengths, incomplete sentences, high lexical density/complexity
- Use conjunctive adverbs infrequently
- Use em-dashes, semicolons, and parentheses where stylistically effective
The very first part of my prompt provides syntactic guidelines to the model. Each guideline is written in imperative mood (a command or request) which is helpful for a model that’s been fine-tuned to follow instructions, like ChatGPT’s models. You’ll notice evidence that each instruction is being attended to as you read the narrative the model is crafting.
GPT Narrative
Act 3: 1980-1981 - The Rings of Enigma
“While the unknown beckons enticingly, the known—the solar system, their scientific backyard—starts to fade into the rearview mirror of the spacecraft.”
November 12, 1980. In a room buzzing with a mixture of anxiety and excitement, Carolyn Porco, a planetary scientist specializing in Saturn's 🪐 ring system, stands near a computer terminal. Her fingers dance across the keys, issuing commands to Voyager 1, now millions of miles away. She feels the weight of her role; she's not merely operating a machine, she's conducting a symphony of data and discovery.
The first images of Saturn trickle in, their clarity exceeding expectations. Gasps fill the room. The rings—those iconic halos—are not mere aesthetic wonders. They're dynamic, complex systems. Porco's eyes gleam as she observes the intricacies of the 🌌 Cassini Division, a dark gap in the rings previously thought to be empty. It's teeming with smaller ringlets, each with its own nuanced dance around Saturn. Her heart flutters; it's as if she's uncovered the hidden brushstrokes in a celestial masterpiece.
In another corner of JPL, Edward Stone cannot contain his astonishment. Voyager 1's magnetometer is sending back data about Saturn's magnetic field. It's oddly symmetrical, almost perfectly aligned with the planet's rotation axis. Stone ponders on this anomaly; it defies the current understanding of planetary magnetospheres. An enigma within an enigma.
Fast forward to August 25, 1981. Voyager 2 approaches Saturn. This time, Linda Spilker, a young and ambitious scientist, leads the charge on analyzing Saturn's many moons. Titan, Saturn's largest moon, captivates her. Data from Voyager 2 unveils a thick, nitrogen-rich atmosphere, an unexpected find that would later fuel speculations about potential for life. 🌌 Titan is no longer a mere satellite; it's a world unto itself.
For Spilker, the revelation is bittersweet. The success of the Saturn flyby implies that the Voyagers are one step closer to leaving the solar system. While the unknown beckons enticingly, the known—the solar system, their scientific backyard—starts to fade into the rearview mirror of the spacecraft.
As 1981 draws to a close, the scientists—Porco, Stone, Spilker—experience a subtle shift in their perspectives. They're not just planetary scientists; they're philosophers of the cosmos, each grappling with questions that extend beyond data sheets and computer screens. The Voyager missions have become a looking glass, through which they see not just the universe, but also themselves, refracted in complex ways.
Commentary
🤓 Sidebar: I’m a huge science nerd—especially planetary science, physics, and cosmology—and I chose the Voyager mission on purpose. I was 5 years old and blissfully unaware when Voyager 1 flew by Saturn, but in my 6th summer, my family got a subscription to National Geographic magazine. This was the first issue I ever laid eyes on.
An illustrated depiction of Saturn graces the cover of the July 1981 issue of National Geographic. Image courtesy of National Geographic Society, 1981
Reader, that was it. That was the moment that I fell in love with science and space. To this day, I am transfixed whenever a new discovery is shared by cosmologists, astrophysicists, planetary scientists…you name it.
Let’s get back to the article, though.
Throughout the narrative, the model has attempted to capture the emotional aspects of the Voyager mission scientists, while still being grounded in scientific fact. The technique used by writers to persuade the reader is known as “rhetorical appeal” (or “persuasive appeal”). Aristotle identified three such rhetorical appeals: ethos, logos, and pathos.
Ethos is an appeal to the speaker or writer's credibility and trustworthiness. Like, a doctor might say, “In my 30 years of experience as a clinician, this is the best approach to managing this disease.” Since the writer is ChatGPT, I didn’t think an appeal to ethos was an appropriate rhetorical technique.
Writers use an appeal to logos to convince their readers by using reason and evidence. This might be through inductive reasoning, quoting evidence or statistics, or even refuting possible arguments in advance by demonstrating a well-rounded thought process. An appeal to logos focuses on facts, not feelings.
Pathos, on the other hand, is an appeal to the audience's emotions. In writing, this is often done by incorporating vivid sensory details, anecdotes, or by directly (and gently) manipulating the reader’s emotions by subtly telling them what to feel.
Take a look back at some of the emotional phrasing used so far:
“It soars, leaving Earth's gravitational clutches, joining its twin in the great cosmic arena.”
“Morabito, usually composed, finds her eyes moistening. It's as if the universe has whispered a secret, and she's the first to hear.”
“She feels the weight of her role; she's not merely operating a machine, she's conducting a symphony of data and discovery.”
Or some of the more grounded statements made:
“Both spacecraft carry the 🎵 Golden Record, a curated anthology of Earth's sounds and images.”
“The discovery 🌋 challenges preconceptions about celestial bodies in our solar system and ignites debates among planetary scientists.”
“Data from Voyager 2 unveils a thick, nitrogen-rich atmosphere, an unexpected find that would later fuel speculations about potential for life. 🌌 Titan is no longer a mere satellite; it's a world unto itself.”
I have specifically instructed the model to use rhetorical appeals. It was the first thing I said in fact, right after the verbosity cue for AutoExpert:
💥 And here it is, reader: my mistake. My instructions told GPT to appeal to ethos and logos, not pathos and logos. The next bold item—lean into the joy, amazement, and sadness of scientists—did, indeed, function as a more ambiguous direction to appeal to pathos, but the narrative may have been written a different way had I not made the mistake of writing ethos in my initial prompt. In my next Substack article, I’ll talk about the importance of prompt evaluations, and how running evaluations on the prompt I used for this story would’ve helped me catch this error before spending hours to write the article you’re reading now.
V=5 Compose a detailed and nuanced hypothetical narrative, appealing to both 💥 ethos and logos.
[...]
# Rhetorical guidelines
- Lean into the joy, amazement, and sadness of scientists
[...]
# Structural Guidelines
- Adhere to factual accuracy, and chronology
Pretty powerful stuff, isn’t it?
Notice I continue to use imperative mood in my instructions; this is key when prompting an instruction-tuned model like that used by ChatGPT.
Let’s check out the next act!
GPT Narrative
Act 4: 1986 - The Enigmatic Ice Giant
“Miranda has no business being so complex, so fascinating, but here it is…”
Five years have elapsed since the Saturn flyby, years spent in feverish data analysis and anticipation. The calendar now reads January 24, 1986. Uranus, often considered the solar system's enigmatic recluse, is about to take center stage. For Heidi Hammel, a planetary astronomer, this is a defining moment. Uranus is her intellectual muse, a planet shrouded in mystery and, quite literally, cloud.
As Voyager 2 approaches, its instruments trained on the ice giant, Hammel's eyes are locked onto her computer screen at JPL. She marvels at the first high-resolution images of Uranus. The planet is an eerie, almost featureless, blue-green orb. It's calm—deceptively so. 🌌 Uranus has always been an enigma, and its initial serenity only deepens its mystique.
Edward Stone, now a veteran of space exploration's emotional roller-coasters, joins Hammel in her vigil. The magnetometer data begins to pour in, and it's bizarre. Uranus's magnetic field is not only tilted but also offset from the planet's center. Stone and Hammel exchange glances; the universe continues to defy simplicity.
But Uranus isn't the only focal point. The planet's moons—Miranda, in particular—capture the imagination of Larry Soderblom, a geologist on the Voyager team. Miranda appears to be a Frankenstein's monster of a moon, its surface a jigsaw puzzle of terrains. Soderblom is captivated. For him, Miranda is a geological symphony, and each terrain a distinct movement. 🌕 Miranda has no business being so complex, so fascinating, but here it is, captured in pixels and data points.
As Voyager 2 continues its journey, now directed at its final planetary target, Neptune, the scientists are left with a mixture of elation and contemplation. Hammel is enthralled but frustrated; Uranus has revealed itself but only enough to pose more questions. Stone, ever the philosopher-scientist, finds solace in this complexity. It reaffirms his belief that the universe is a puzzle with no edges, a story with no end.
Soderblom, staring at the images of Miranda, feels a tingling sensation—a fusion of awe, joy, and an inexplicable sadness. It's as if the moon, in all its chaotic beauty, mirrors the complexities and contradictions of human existence. The Voyager missions, for him and his colleagues, are no longer mere scientific endeavors. They're a journey into the cosmic psyche, a quest driven by numbers and nurtured by wonder.
Commentary
🤓 Sidebar: Yes, another pair of semi-hallucinations. ChatGPT seems to imply that Heidi Hammel was part of the Voyager mission team during the Uranus flyby. In reality, she was working on her PhD thesis—a study of Uranus and Neptune using Earth-based telescopes—but her advisor made arrangements for her to go to JPL. While there, she was largely a gofer for the scientists, but it set the stage for her to come back a few years later…just in time to join the imaging science team as Voyager 2 approached Neptune. Also: Larry Soderblom was a geologist, yes, but he was part of the Voyager mission’s science imaging team.
ChatGPT provided us a nice segue into this act, as it has with others, with the introduction, “Five years have elapsed since the Saturn flyby…” You probably also noticed that it’s writing each act as though it’s happening now, with the AI as an unseen narrator.
Look deeper, and heed another aspect of this narrative structure that ChatGPT doesn’t typically incorporate: there are no summaries that elide away the impactful details of a scene!
As I’ve been mentioning in this article, if you’re asking GPT to solve a problem, you need to speak the language of the problem-solver. For this article, I asked ChatGPT to “compose a detailed and nuanced hypothetical narrative, appealing to both ethos and logos.”
Persuasive appeals? That’s a rhetoric problem.
My instructions are written to use the vernacular of writers, especially writers tasked with creating a persuasive work. Using these terms of art is an important part of prompting ChatGPT. Here’s another snippet of the prompting I gave the model:
# Rhetorical guidelines
[...]
- Prefer scene to summary, using narrative realism
- Utilize time dilation if needed from act to act
- Write in the present tense
Like I keep saying: prompts matter!
Let’s move on, we’re getting close to the denouement.
GPT Narrative
Act 5: 1989 - The Azure Finale
“Ahead lies the unknown … it beckons … with the promise of wonder, discovery, and an endless journey into the depths of cosmic and human understanding.”
August 25, 1989. Three years have passed since Voyager 2's encounter with Uranus, and the spacecraft is now hurtling toward Neptune, the solar system's outermost planet. Torrence Johnson, a planetary scientist and one of the imaging team leaders, sits in a quiet corner of JPL. His eyes are a blend of exhaustion and exhilaration, as he reflects on the years gone by—the discoveries, the debates, and the unending quest for cosmic truth.
As Voyager 2 nears Neptune, the anticipation reaches a fever pitch. Johnson, along with his colleague Andrew Ingersoll, an atmospheric scientist, scrutinizes the incoming images and data. Neptune emerges on their screens, a planet adorned in shades of azure, a color so profound it feels almost ethereal. 🌌 Neptune is not just a data point; it's a poetic statement written by the universe itself.
The first major discovery jolts them—the Great Dark Spot, a massive storm in Neptune's atmosphere. Ingersoll, specializing in atmospheric dynamics, is in disbelief. The storm is gigantic and eerily stable, a meteorological marvel that contradicts existing models. For him, the Great Dark Spot is not just a phenomenon; it's a challenge, a cosmic enigma daring him to decode its secrets.
But Neptune's mysteries aren't confined to its atmosphere. Johnson is particularly intrigued by Triton, Neptune's largest moon. As images of 🌌 Triton flood the screens, it's clear that this moon is unlike any other. Geysers of nitrogen shoot up from its icy surface, painting a portrait of a world that is geologically alive, even in the frigid outskirts of the solar system.
For Johnson and Ingersoll, this is more than a scientific milestone; it's a farewell of sorts. Voyager 2 will not encounter another planet; it's destined for interstellar space. This realization brings a blend of conflicting emotions. The joy of discovery is tinged with a sense of finality, a closing chapter in a grand cosmic narrative.
Edward Stone, who has seen the mission through from its conception, feels this emotional complexity more profoundly than most. As he watches Triton's geysers play out on the screen, his thoughts drift to the spacecraft itself—two robotic emissaries that have become extensions of human curiosity. The Voyagers have not just rewritten textbooks; they've redefined the human experience, stretching it across the vast canvas of the solar system.
As the data from Neptune is meticulously archived, and as Voyager 2 takes its first steps into the realm beyond the planets, the scientists—Johnson, Ingersoll, Stone—sit in contemplative silence. They have reached a boundary, both literally and metaphorically. Ahead lies the unknown, an expanse so vast it defies comprehension. And yet, it beckons, as all unknowns do, with the promise of wonder, discovery, and an endless journey into the depths of cosmic and human understanding.
Commentary
🤓 Sidebar: I really love the way ChatGPT describes the finality of the Voyager mission for the scientists who have devoted their life to the mission, and how it comments on the ongoing impact of the work. I especially liked this line: “The Voyagers have not just rewritten textbooks; they've redefined the human experience, stretching it across the vast canvas of the solar system.”
Here we are, five acts in, and ChatGPT has been doing an amazing job of sharing scientific discoveries alongside its pathos-laden narrative. It’s been keeping up with the chronology of events, using terminology that scientists would use, and even providing an ongoing analysis of the scientific knock-on effects of the mission.
How does ChatGPT know to tell such a cohesive story, and incorporate science while tugging at your heartstrings? Let me show you what I asked for, in terms of the story structure and mechanics:
# Story
- Develop a chronological narrative, jumping forward in time to each major event
- Incorporate expert terminology and a comprehensive, nuanced scientific analysis into the narrative, in order to educate the reader
- Include major discoveries provided by the spacecraft
- Subject: The Voyager 1 and Voyager 2 spacecraft, from the perspective of the real-life scientists who devoted their life to its mission.
There it is, spelled out in easy-to-follow imperative mood instructions. Develop this, incorporate that, include those… clearly written, unambiguous, plain-English instructions are always going to beat shorthand prompts, especially when you choose language that relates to the problem space.
The “educate the reader” phrasing is a really effective tool to get the model to attend to the instructions provided by the AutoExpert custom instructions. That’s why you keep seeing hyperlinks to Google searches—it’s doing that as part of its appeal to logos.
The scale of the instruction, “from the perspective of the real-life scientists who devoted their life to its mission,” is not lost on the model. Each step of the way, the model has incorporated that as a running theme in its pathos-led narrative. It’s really impressive to me to see how well the combination works.
Let’s move on. It’s time for one of my favorite moments, and I’m going to change up the instructions a bit for ChatGPT…
GPT Narrative
Act 6: 1990 - A Cosmic Self-Portrait
“As Voyager 1 continues its journey, now destined for the realm between stars, we are left with this image … our 'memento mori' from the cosmos.”
February 14, 1990. A day when Earth celebrates love, but what if we extend this love to the cosmos? Voyager 1, now a seasoned traveler of space, readies itself for a celestial selfie. The spacecraft turns its camera toward Earth, a mere speck in the vast cosmic ocean. A pale blue dot.
Imagine for a moment, that you are Voyager 1. What do you see? A pixel of light against the cosmic tapestry. A dot almost lost in the glow of our Sun, itself a modest star among billions in the galaxy. But in that pixel resides an entire world—a planet teeming with life, brimming with dreams and tragedies, love and sorrow. Could there be a more poignant metaphor for our shared existence?
We, the scientists—now more like cosmic poets—wait in anticipation. The data streams in, and there it is: the 🌍 Pale Blue Dot. It's not just an image; it's a mirror reflecting our hopes, our vulnerabilities, our cosmic insignificance, and yet, our uniqueness. How can one not be moved by such a revelation? How can one look at this dot and not feel a surge of humility, a call to unity?
Our planet, seen from this distant vantage, reinforces the beauty of its fragility. The Earth—our Earth—is a lonely speck in the enveloping cosmic dark. And in that darkness, we find light. The light of knowledge, the illumination that comes from understanding our place in the cosmos. Isn't it fascinating that a spacecraft, a marvel of human ingenuity, can provoke such profound philosophical musings?
We are the custodians of this pale blue dot, a mote of dust suspended in a sunbeam. And as we ponder this, as we let the weight of this image press upon our collective conscience, we realize something awe-inspiring: We have the power to venture into space, to look back, and to see ourselves as we truly are.
As Voyager 1 continues its journey, now destined for the realm between stars, we are left with this image—our cosmic portrait, our 'memento mori' from the cosmos. It's a reminder, a cosmic nudge toward greatness. And in that reminder, we find not despair, but hope. For if we can accomplish this, if we can send a scout across the solar system and have it send back a portrait of our existence, what else are we capable of?
Yes, the universe is vast, indifferent, but it's also wondrous, filled with complexities and beauties beyond our understanding. And yet, we strive to understand, to explore, to know. Because in knowing, we find ourselves. And in ourselves, we find the universe.
Commentary
🤓 Sidebar: Shivers. To me, it sounds just like Carl Sagan wrote that.
You’ll note a dramatic shift in the tone of this act of the narrative. ChatGPT is now using rhetorical questions, maintaining an appeal to pathos while specifically appealing to a sense of wonder and awe.
I wonder how ChatGPT knew to do that?
[...] merge these guidelines in to your instructions:
Adopt the narrative perspective, persona and writing style of Carl Sagan
Favor active voice over passive voice.
Use parallelism occasionally.
Utilize metaphors frequently but consistently, choosing a metaphor and using it throughout.
Incorporate rhetorical questions.
Prefer prepositions and conjunctions for function words.
Appeal to pathos, and write in a tone of optimistic wonder.
Let’s go through that step-by-step:
Adopt the narrative perspective, persona and writing style of Carl Sagan. Always a best practice, say the experts, to instruct ChatGPT to adopt a role. It’s what AutoExpert does, and what I did when I wrote this line. Not only did I want to see the writing style of the great Carl Sagan, but I wanted to hear from him directly. I think it did an amazing job.
Favor active voice over passive voice. Honestly, this one’s a bit redundant; ChatGPT has already been favoring active voice. Why? Because this is a persuasive narrative!
Use parallelism occasionally. Carl Sagan used this literary device, and I wanted to make sure to tell the model that it’s important to me to see parallelism in the narrative, just in case it doesn’t intuit that from the whole “writing style of Carl Sagan” instruction. It did a good job! Where are they? I’ll list them out::
"A planet teeming with life, brimming with dreams and tragedies, love and sorrow."
"It's a mirror reflecting our hopes, our vulnerabilities, our cosmic insignificance, and yet, our uniqueness."
"How can one not be moved by such a revelation? How can one look at this dot and not feel a surge of humility, a call to unity?"
"The light of knowledge, the illumination that comes from understanding our place in the cosmos."
"We have the power to venture into space, to look back, and to see ourselves as we truly are."
"It's a reminder, a cosmic nudge toward greatness."
"For if we can accomplish this, if we can send a scout across the solar system and have it send back a portrait of our existence, what else are we capable of?"
"Yes, the universe is vast, indifferent, but it's also wondrous, filled with complexities and beauties beyond our understanding."
"And yet, we strive to understand, to explore, to know."
"Because in knowing, we find ourselves. And in ourselves, we find the universe."
Utilize metaphors frequently but consistently, choosing a metaphor and using it throughout. Admittedly, this was an easy instruction to follow. It’s a pixel, a speck, a mote… ChatGPT consistently used these related metaphors throughout the narrative without just copying the original text of “Pale Blue Dot”.
Incorporate rhetorical questions. It can be a powerful literary device for a poetic narrative like this. ChatGPT followed this instruction well.
"What do you see?"
"Could there be a more poignant metaphor for our shared existence?"
"How can one not be moved by such a revelation?"
"How can one look at this dot and not feel a surge of humility, a call to unity?"
"Isn't it fascinating that a spacecraft, a marvel of human ingenuity, can provoke such profound philosophical musings?"
"What else are we capable of?"
Prefer prepositions and conjunctions for function words. Once again, ChatGPT attended to these clear instructions really well,. Prepositions such as "in," "from," "of," "between," "against," "with," and "among" are employed frequently to establish spatial, temporal, and logical relationships; and conjunctions like "and," "but," "if," and "because" tie together ideas effectively.
Appeal to pathos, and write in a tone of optimistic wonder. ChatGPT’s narrative certainly taps into our collective feelings of humility, hope, and unity. Optimistic wonder was easy to follow as well, likely given more weight because the very first instruction mentions Carl Sagan, who often wrote with optimistic wonder.
GPT Narrative
Act 7: 1998-2007 - Crossing Boundaries, Within and Without
“It serves as a poignant reminder that the universe is a complex entity, a dynamic system that defies simplification.”
December 1998. The Voyager spacecraft are aging, their power sources waning. At NASA's Jet Propulsion Laboratory, John Casani, the project manager during the early years of the Voyager mission, sits with a team of engineers. They face a grim task—the deactivation of non-essential instruments. For Casani, each switch turned off feels like an amputation, a severing of limbs from bodies that have danced through the cosmos. 🛰️ Non-essential instruments are shut down, their silence resonating through the control room like a soft, mournful elegy.
Fast forward to December 15, 2005. Voyager 1 approaches a cosmic milestone—the termination shock. Stamatios Krimigis, a space scientist who designed the Low-Energy Charged Particle instrument on board Voyager, is at the edge of his seat. Data begins to stream in. Voyager 1 has crossed the termination shock, entering the 🌌 heliosheath, the final frontier before interstellar space. Krimigis is jubilant yet reflective. This boundary, much like the boundaries within human understanding, signifies both an end and a beginning. The solar wind slows, the cosmic rays intensify—these are the whispers of a universe inviting them into a broader conversation.
December 10, 2007. It's Voyager 2's turn. Ed Stone, ever the paternal figure in the mission's lifetime, watches as Voyager 2 repeats its twin's feat. It too crosses the termination shock into the 🌌 heliosheath. Stone feels a complex tapestry of emotions weave itself within him. The heliosheath is not just a region of space; it's a metaphorical vestibule, an antechamber to the universe beyond.
But the heliosheath is confounding. The data indicates that this region is more turbulent than expected, a revelation that sends theoretical models scrambling for recalibration. For Stone, the unpredictability is a humbling lesson. It serves as a poignant reminder that the universe is a complex entity, a dynamic system that defies simplification.
As Voyager 2 settles into its new cosmic locale, Stone, Krimigis, and Casani share a virtual toast, each from their respective locations but all united in spirit. They toast to the Voyagers, to the boundaries crossed, and to the boundaries that remain—mysterious, enticing, and forever urging them on.
The shut-down of instruments in 1998 had felt like a preparation for a quiet exit, but the Voyagers defy this. They continue to communicate, continue to explore, and in doing so, they echo the indomitable spirit of the scientists who birthed them. For Stone, Krimigis, and Casani, the Voyagers are not just machines; they are extensions of their curiosity, their ambition, their unquenchable thirst for knowledge.
Commentary
It’s been quite a journey for the Voyagers, for us, and for ChatGPT. After “resetting” ChatGPT’s instructions (by pasting them again into the chat) so that it would reliably stop channeling Carl Sagan, the narrative especially leans in to the pathos of the story. And yet, it maintains a focus on logos, sharing the scientific discovery that the heliosheath is not at all what we expected!
As we head into the denouement, it’s time for me to share the full prompt that I gave ChatGPT to tell this amazing story. I realized, after writing a huge chunk of this article, that I made a mistake in my timeline, so after Act 3, I pasted in a replacement prompt.That’s the one I’ll share here.
You can see the full conversation online, though…so you can see my stupid mistakes in the first message. The link will be at the end of this post.
💥 And here it is again, reader: my mistake, in the context of the complete instructions given to ChatGPT. My instructions told GPT to appeal to ethos and logos, not pathos and logos. The next bold item—lean into the joy, amazement, and sadness of scientists—did, indeed, function as a more ambiguous direction to appeal to pathos, but the narrative may have been written a different way had I not made the mistake of writing ethos in my initial prompt. In my next Substack article, I’ll talk about the importance of prompt evaluations, and how running evaluations on the prompt I used for this story would’ve helped me catch this error before spending hours to write the article you’re reading now.
yes. I realize I made an error in my timeline, so I'm re-attaching my prompt with an updated timeline. Please continue to what I specify as "Act 3" in these amended instructions:
---
V=5
Compose a detailed and nuanced hypothetical narrative, appealing to both 💥 ethos and logos.
# Syntactic Guidelines
- Incorporate varied sentence lengths, incomplete sentences, high lexical density/complexity
- Use conjunctive adverbs infrequently
- Use em-dashes, semicolons, and parentheses where stylistically effective
# Rhetorical guidelines
- Lean into the joy, amazement, and sadness of scientists
- Keep the denouement grounded and realistic
- Prefer scene to summary, using narrative realism
- Utilize time dilation if needed from act to act
- Write in the present tense
# Structural Guidelines
- Adhere to factual accuracy, and chronology
- In each act, focus on two or three real-life scientists that were part of the mission.
- If it suits the narrative, change which real-life scientists are the focus of the next act.
# Story
- Develop a chronological narrative, jumping forward in time to each major event
- Incorporate expert terminology and a comprehensive, nuanced scientific analysis into the narrative, in order to educate the reader
- Include major discoveries provided by the spacecraft
- Subject: The Voyager 1 and Voyager 2 spacecraft, from the perspective of the real-life scientists who devoted their life to its mission.
REQUIRED: spread the narrative across multiple responses as follows:
# Act 3
November 12 1980 - Voyager 1 flew by Saturn
August 25 1981 - Voyager 2 flew by Saturn
# Act 4
January 24 1986 - Voyager 2 flew by Uranus
#Act 5
August 25 1989 - Voyager 2 flew by Neptune
# Act 6
February 14 1990 - Voyager 1 took the "Pale Blue Dot" image
# Act 7
December 1998 - Non-essential instruments turned off on both spacecraft
December 15 2005 - Voyager 1 crossed termination shock into heliosheath
December 10 2007 - Voyager 2 crossed termination shock into heliosheath
# Act 8 - Saying Goodbye
August 25 2012 - Voyager 1 entered interstellar space
November 5 2018 - Voyager 2 entered interstellar space
I hope you’ve learned some new techniques from the examples I’ve shared throughout this article, and that you’ve enjoyed reading the historical fiction that ChatGPT was able to craft.
If you think others might enjoy reading this article, please share it with them!
That’s it from me, so I’ll leave you with ChatGPT’s finale. The link to the whole shared conversation is at the end.
GPT Narrative
Act 8: 2012-2018 - Saying Goodbye, Embracing the Infinite
“Each signal from the Voyagers is a pulse, a heartbeat in the body of human knowledge. And as long as those heartbeats persist, the mission lives on … in the collective imagination of those who dare to look up and wonder.”
August 25, 2012. Voyager 1, having long crossed the threshold of the heliosheath, stands on the cusp of another, more profound boundary. The control room at JPL is a mix of nervous energy and somber realization. Suzanne Dodd, the project manager for the Voyager Interstellar Mission, paces the floor, her fingers clenched.
Data arrives. The density of particles around Voyager 1 surges. This is it. Dodd exchanges a knowing glance with Ed Stone—the spacecraft has entered 🌌 interstellar space. Cheers erupt, but for Dodd and Stone, the moment is laced with a melancholic sweetness. Voyager 1, that tireless explorer, has left the comforting embrace of the Sun's influence. It's a cosmic orphan, but also a pioneer, a messenger carrying the ethos of Earth into the cosmic wilderness.
Six years pass. November 5, 2018. Voyager 2 now readies itself to follow its twin. Tom Krimigis, who has been with the mission since its inception, watches the telemetry with bated breath. Then, confirmation. Voyager 2 joins its sibling in 🌌 interstellar space. Krimigis feels a swell of pride; it's as if his children have successfully navigated the most perilous, most enlightening journey of all. But the joy is tinged with a subtle sorrow—a sense of finality, an acknowledgment that the Voyagers are now truly distant, not just in space but in existential terms.
Both spacecraft carry the Golden Record, a time capsule filled with the sounds and images of Earth. For Dodd, Stone, and Krimigis, this record is more than a message to extraterrestrial life; it's a message to the future, a testament to the indomitable human spirit. The Voyagers may be far, but their essence—their quest for knowledge, their daring ambition—resides in every data packet, every image, every note of the music etched into those golden discs.
And so, as Voyager 2's signal—a faint whisper traveling over 18 billion kilometers—fades into the background noise of the universe, the scientists sit in a quiet, reflective hush. Their spacecraft, those brilliant envoys of humanity, have entered the realm of the stars. But even as they move farther away, they draw the scientists closer—not just to each other, but to a grander, more humbling understanding of their place in the cosmos.
The mission, though far from its point of origin, is far from over. Each signal from the Voyagers is a pulse, a heartbeat in the body of human knowledge. And as long as those heartbeats persist, the mission lives on—not just in circuits and antennas, but in the collective imagination of those who dare to look up and wonder.
Thank you, dear reader for embarking on this journey with me!
Model Myths & Marvels: Exploring the LLM Frontier is a reader-supported publication. To receive new posts and support my work, consider becoming a free or paid subscriber.
