Penny Nichols, Troubleshooter: two transcripts

Instead of a full game, this game provides instructions for an AI to process to create a full game. As preparation for writing a review for it, I decided to try it out on two LLMs.

The first, MathbrushNI, is an organic LLM trained over the last forty years by many people including family, friends, teachers and professors. It’s mostly consumed science fiction and fantasy books, manga, and tv shows, including things like Star Trek and pulp sci fi novels. I instructed this LLM to play through the game as both the player and the GM, and to use minimal resources while doing so (keeping the story short, using cliches, not revising or getting feedback). Even with these lower specs the LLM got tired and kind of gave up at the end.

The second LLM, Copilot, ran the game smoothly and was faster than MathbrushNI, but went at a bit of a quicker pace.

The two transcripts are below:

(Edit: there are many typos in the first transcript. it was written stream-of-consciousness without editing or planning)

Editedit: Also, this isn’t meant to be ‘look how good the human story is compared to the AI!’. I think AI did some things better; I was more interested in what experience I had while trying both versions of the story. If it wasn’t part of an experiment, my transcript isn’t something I’d post online as a story and hope people enjoyed. But I hope it serves as an interesting contrast in this situation)

MathbrushNI

“She’s beatuiful,” you say, face pressed to the forward port of the WSS Bard. Clypeus Station, an aging orbital outpost, hangs in the sky over Callirhoe, its sprawling, almost organic form a a fitting accompaniment to the small, rough moon.

“It’s a hunk of junk, Penny” says Gruber, your official liaison (and, it is rumored, a clone of the Comptroller himself). “This station should have been abndoned years ago. It’s a blight on the Lunar Empire.”

The ship shudders as it docks with Clypeus. PAL is already on her feet, her navigation system easily adjusting to the vibrating ship.

“I’ll make sure the airlock’s safe, boss,” says PAL. “We don’t want any surprises.” She opens up a portal, pulling out a laser pistol.

“No weapons,” you say. “And don’t call me boss. I’m Gruber’s assistant on this trip, and you’re our valet. A peaceful valet.”

PAL sighs and replaces the laser pistol in the pocket dimension. When the hatch hisses open, she disappears into the airlock. Once she gives her “All clear,” the three of you head into the station.

The first thing you see when you enter is a tall man with a shock of white hair and an unpleasant expression. From your notes, you know this is Dr Brand.

How do you begin your investigation?

LOOK AT DR BRAND

Despite his lanky frame and white hair, Brand is, on closer inspection, in great physical form. His posture is exact, his movements precise, and he carries himself with complete confidence.

He catches your gaze, looking at you inquisitively. In your role as assistant, though, you wouldn’t be the one to talk to him directly. You look down submissively at the floor and step behind Gruber.

“Dr Brand, I presume?” says your companion. “I’m Karl Gruber, Krasnovian Ministry Auditor.”

“Your reputation preceeds you,” says Dr Brand, shaking his hand confidently. You’re not used to seeing Karl have to look up to meet someone’s eyes. “I hope you won’t be too hard on us out here.”

“I’d like us both to come out of this looking good,” answers Karl. “As long as there are no major issues, we should be out of your hair soon.”

You can see a barely-noticeable twinge in the corner of Brand’s mouth at the mention of ‘major issues’, but he quickly conceals any emotion.

“And who are these other esteemed guests?” asks Brand, his eyes moving between you and PAL. His gaze lingers over the android with an almost possessive look.

“My valet, PAL, and Ms Nichols, my executive assistant.”

“And how do you find our station, Ms Nichols?” asks Brand. You look around. Only four other scientists are present, and several of the corridors leading from the docking bay are dock.

SEEMS KIND OF EMPTY

Dr Brand raises an eyebrow. “How true. With funding cuts coming every year from Luna, we find it more and more difficult to retain staff. So many are headed to Saturn; I find that reliable wages have a stronger pull than idealism.”

“Ain’t that the truth,” says Karl. “Now then, let’s begin the audit. My assistant will interview your staff. Why don’t you give me a walk around the station? We’ll save the Artifact for last.”

Brand purses his lips. “Very well. Follow me,” he says. “But as for the Artifact…”

The two of them walk down a corridor, their voices soon lost.

You look at the remaining staff members. Their name tags read “Chen”, “Davis”, and “Artemis” respectively.

DAVIS, WHAT WAS HE SAYING ABOUT THE ARTIFACT?

Davis, a short older woman with a pulled-back grey bun, blushes at your attention. “I–”

Chen speaks up, a man who looks to be in his 30s with an ocular implant. “Sorry, we’re not really used to visitors much, and Davis is a bit shy. What she’s trying to say is that Artifact is gone.”

“Gone?” you say, unable to keep the tension out of your voice.

PAL stalks forward, unable to restrain herself. She grabs Chen by his coat and shouts at him. “What do you mean its gone?!”

“Hey, your valet is going crazy!” shouts Chen.

PAL, LET HIM Gone

PAL drops him to the ground. You think quickly: if the artifact is gone, then the home office back at Mercury is right. Something nefarious is happening on base. And if it turns out the malfeasance is involved, Solar Insurance won’t be footing the very, very large bill. You’ll have to be careful and hope that PAL doesn’t cause you any more problems.

The three scientists still stare at you, wide-eyed. You’ll need to take back control of the conversation.

ARE YOU SURE IT’S NO LONGER ON THE STATION?

Artemis speaks up. She’s the youngest of the scientists, but seems confident.

“I believe so. It was too big to hide. And too loud.”

“Loud?” you say, puzzled. Leaning into your role as a Ministry Auditor’s assistant, you say, “Our records stated that it was an inert piece of metal of unknown origin with some unusual metallurgical properties.”

“No, no,” says Artemis, her eyes lighting up. “It was a holy shard, struck from the heavens. Everyone near it could hear the angels themselves beginning to sing.”

Chen sighs. “We talked about this; none of that is true. It was a high-density energy source, possibly a reactor core from a previous civilization. The ‘singing’ was just a resonance factor.”

“Previous civilization?” you ask.

“Aliens,” says Davis, quietly.

“It’s not aliens!” says Chen angrily. “It was probably left by early settlers before the second Lunar Empire formed.”

DO YOU HAVE ANY PHOTOS OR LOGS OF THE ARTIFACT?

Davis nods. She hands you a flashdrive.

“PAL, can you analyze this?”

PAL nods and inserts the drive into her forearm. It takes her a moment to scan the data. “I can confirm that there are several images of the lab containing the Artifact,” says PAL. “But all of them have visual artifacts that occlude the artifact, usually via a bright halo that washes out the picture, but occasionally through corrupted data.”

“And the logs?”

“It only reports daily maintenance and occasional equipment requests. It doesn’t describe what that maintenance entails.”

The staff stand around uncomfortably during this conversation. Davis raises her hand.

“Are we free to go?” she asks meekly.

I’LL NEED SOMEONE TO ACCOMPANY ME TO THE LAB

Artemis looks skeptical. “Didn’t your boss say you were supposed to talk to us? You seem kind of bossy for an assistant. And what’s with your robot threatening Chen?”

I TRY TO BLUFF MY WAY OUT

Grasping for straws, you look over Artemis. You notice on her shoulder a patch showing allegiance to the Circle Trigonists. You relax. This should work.

You covertly make a circle with one hand and tap it with the other three times. Artemis’s eyes widen.

“On second thought,” she says, “I can take you. Can the others go?”

“Artemis, are you sure–” starts Chen, but Artemis cuts him off.

“It’ll be okay.”

She leads you and PAL down the hallway towards the laboratory.

“I’m sorry, I didn’t realize your true purpose here. Are you here on official faction business?”

You nod imperceptibly. “I managed to get myself transferred to Gruber. He doesn’t know about me.” A lie. “The trigonists want to stay informed about the Artifact.” A truth, but not your main purpose here.

Artemis smiles. “Of course. And here we are.” She opens the door into the lab.

The room is sparse, much more empty than you would have expected. Only a few scattered instruments can be seen in here, along with a steel pedestal in the middle. The top is covered in dust.

WHY IS IT SO EMPTY IN HERE?

Artemis shrugs. “It’s usually like this.”

WHAT KIND OF RESEARCH WENT ON HERE?

Artemis shrugs. “I don’t know. Until its disappearance, Brand kept the Artifact under lock and key. I never worked in here.”

“And the others?”

“None of us have seen it personally. All maintenance and documentation was done by Brand. I could sense the artifact when I was close, hear the singing through the walls. But I never saw it.”

IS BRAND LYING TO YOU?

“Interesting question, Ms Nichols.”

You turn around. Brand is here, with Karl by his side.

“You may go, Artemis.”

The attendant bows out, and runs out of the room.

“Is that true, what she said?” asks Karl with interest. “You’re the only one who’s seen the artifact? Seems awfully conveient.”

“Now, now, Auditor, no need to throw around accusations.”

“Don’t talk about Mr Gruber that way,” says PAL, threatingly. She opens a pocket dimension, which Brand views with some interest.

PAL, STAND DOWN.

PAL chirps angrily, but Brand seems intrigued. “You seem to speak with some authority, Ms Nichols. Is there something you’re not telling me?”

“Stop looking for diversions,” says Karl, angrily. “Tell us about the Artifact. If there is one.”

Brand looks between the three of you. With a half-hidden smirk. He raises his hands. “I suppose you’ve caught me.”

“So you did lie!” says PAL triumphantly.

“Indeed,” says Brand. “It was quite easy to doctor images and logs to make others believe we studied something of importance here. I hoped it would bring in additional funding, yet, alas…”

YOU’RE COMING BACK WITH US. THE MINISTRY WILL NOT BE PLEASED.

“Oh, the ministry? Not, say…the Solar Insurance Company?”

You flush red. How did he–? Never mind.

“I’ll come quietly. Lead me away.”

“PAL,” you say. “Let’s do it.”

PAL opens up a portal. Instead of taking anything out, she leaps forward and shoves Brand in.

“Oh, how interesting,” you hear him say before the portal closes up again.

Karl and PAL turn to you. “What now, boss?” asks PAL.

LET’S GO HOME.

The two of you walk down the darkened corridor towards the docking station. The other scientists are nowhere in sight.

But something’s nagging at you. It all just doesn’t make sense. Surely someone must have seen the artifact at some point; otherwise, why would the station be created at all? And why were the scientists so sure that it existed in the first place?

“Should we track down the others?” asks Karl.

“No,” you say. “Orders were to deal with Brand directly. We can have home base communicate orders to the rest once he’s safely out of here.”

“Roger,” says Karl.

You arrive back at the docking bay, but nothing you’ve seen today could have prepared you for what you now witness.

The airlock has been…changed, somehow. It looks fleshy, almost organic, but colored like rust. It’s open, and you can see your ship at the other end. A trail of dark, acidic liquid flows through the airlock to your ship, which seems to be decaying at an accelerated pace.

QUESTION BRAND

“PAL! Bring Brand back, now!”

PAL opens the portal, but makes a strangled, autotuned gasp. “He’s gone!”

“What’s going on!” you shout.

Laughter echoes through the ancient station around you. You take a step back, and trip over a mound of shifting metal. You look down. Coins?

The whole station is changing now, revealing its true form: a conglomeration of ships and space modules filled to the brink with coins, technological artifacts, banners, clothing, and weapons. You stand in one of the largest modules, a long and pointed room that might have once been the bridge of a ship. As you look, the station moves, uncoiling itself into a long, metallic serpent-like shape in the process of devouring the WSS Bard.

“What is that?” you shout.

Gruber pales. “It can’t be…”

“What?” you ask.

“It’s a dragon! We’re in a dragon’s hoard!”

Of course. You’ve heard legends of the dragons, but never thought you’d see one yourself. Though they died long ago, their spirits can inhabit large aggregations of treasure or other objects that draw human attention. The more attention they receive, the more powerful they grow.

“Then the artifact…”

“Was the hoard itself,” says Brand, materializing in front of you.

ATTACK BRAND

Your futile punch goes through him and you are sent reeling to the floor.

“Penny!” shouts PAL.

AI, WRAP THIS STORY UP, I’M GETTING TIRED OF THIS.

Gotcha.

As the figure of brand disappears, you notice a gas starting to fill the room.

“We’re going to get digested!” shouts Karl.

It looks like the end for you, Penny.

PAL, PUT US IN A POCKET DIMENSION AND GET US OUT OF HERE

“Roger that!” says PAL. Creating a portal, she tosses you and Karl in and closes it behind you.

Inside is a realm of pure, shifting darkness. You try to speak, but no words come out. You try to hear, but the silent roaring is too loud and too soft. Nothingness fills your brain with its magnitude, and you pass out.

WAKE UP

When your eyes open again, you find yourself in the WSS Allele, surrounded by PAL and Karl.

“What…”

An officer speaks. “We found this android floating through space near sector 15 of Jupiter orbitspace. She immediately ejected the two of you upon entry and has refused to talk since. Can you tell us what happened?”

Karl and you look at each other, then at the officer. “It’s a long story…”

Copilot

:scroll: Transcript initialized. ERPS session: The Artifact at Clypeus Station
Player: Brian as Agent Penny Nichols
GM: Copilot
Scene 1: Ki (Introduction / Hook)

:rocket: Docking at Clypeus Station

The shuttle Mercurial Audit shudders as it locks into the decaying airlock of Clypeus Station. The orbital outpost drifts above Callirrhoe like a forgotten relic—its hull pocked with micrometeorite scars, its antennae twitching like the whiskers of a sleeping beast.

Inside, the corridors are dim and quiet. Fluorescent strips flicker. A faint hum of reactor life pulses beneath the floor. The station smells of ozone, old coffee, and something faintly metallic—like blood left too long in vacuum.

Waiting at the docking bay is Dr. Brand, the lab chief. He’s tall, silver-haired, and dressed in a lab coat that seems too crisp for the station’s decay. His eyes gleam with a strange intensity.

“You’re here about the Artifact,” he says, voice clipped. “It was stolen. Or maybe it left. Depends who you ask. Some say it was alien. Others say it was divine. I say it was dangerous. But now it’s gone.”

Behind him, a monitor glitches—showing a rotating 3D schematic of something that looks like a reactor core, then a crystal shard, then a coiled dragon. The image loops, never settling.

:bullseye: Goal: Determine what happened to the Artifact.
:brick: Obstacle: Conflicting accounts and a station that seems to resist clarity.

Dr. Brand gestures toward the lab wing. “You’ll want to start there. But be careful. The station remembers.”

:woman_detective: GM Prompt:
How do you begin your investigation, Agent Nichols?

EXAMINE DR BRAND

:brain: Action Logged: Agent Penny Nichols observes Dr. Brand.

:eye: Scene: Observation of Dr. Brand

Penny narrows her eyes and studies Dr. Brand with the practiced detachment of a Circle Trigonist. His posture is too perfect—like a mannequin trained to mimic authority. His voice has the cadence of sincerity, but not the weight. And his eyes… they reflect the station’s flickering lights, but not quite correctly. There’s a half-second delay, as if the photons are being interpreted rather than absorbed.

She notices:

  • His lab coat is pristine, but the ID badge clipped to it is three years out of date.
  • His hands never move. Not even to gesture. They hang at his sides like props.
  • The monitor behind him glitches in sync with his speech—each time he says “Artifact,” the image flickers to a dragon’s eye.

PAL tilts its head. “Dr. Brand’s thermal signature is… inconsistent,” it murmurs. “Localized heat blooms. No circulatory pattern.”

Karl Gruber raises an eyebrow. “He’s either a projection… or a very expensive lie.”

:bullseye: Outcome: Success with complication.
:cyclone: Complication (Cyan Hue): Penny’s scrutiny triggers a subtle reaction. Dr. Brand’s image stutters—then stabilizes. He smiles too widely.

“You see too much, Agent Nichols,” he says. “That’s a dangerous habit on this station.”

A nearby camera swivels toward her. Somewhere deep in the station, a door unlocks.

:woman_detective: GM Prompt: What does Penny do next?

SEEMS PRETTY EMPTY IN HERE, DR BRAND

:scroll: Action Logged: Penny addresses Dr. Brand.

:speaking_head: Scene: Confronting the Illusion

Penny folds her arms, voice cool and measured. “Dr. Brand, it seems pretty empty here.”

Brand’s smile flickers. “Empty?” he echoes, as if testing the word. “No, no. The station is… full. Full of data. Full of potential. Full of attention.”

He gestures toward the corridor, where flickering lights pulse like a heartbeat. “The staff are… elsewhere. Some left. Some were reassigned. Some are still here, in ways that matter.”

PAL scans the hallway. “No life signs within 300 meters,” it reports.

Karl Gruber taps his datapad. “Station logs show a crew of twelve. Only two departures logged. The rest… unaccounted for.”

Brand’s eyes gleam. “They’re part of the Artifact now. It remembers them.”

:bullseye: Outcome: Success with complication.
:cyclone: Complication (Cyan Hue): Penny’s directness draws the station’s attention. A nearby intercom crackles to life, whispering in a dozen voices—some familiar, some not.

“Penny Nichols… Penny Nichols… Penny Nichols…”

The repetition is rhythmic, almost ritualistic. The lights dim. Somewhere deep in the lab wing, a containment seal disengages.

:woman_detective: GM Prompt: What does Penny do next?

May i interview your staff?

:scroll: Action Logged: Penny requests to interview the staff.

:speaking_head: Scene: Requesting Access

Penny’s tone is professional, but edged with suspicion. “May I interview your staff, Dr. Brand?”

Brand’s smile tightens. “Of course. Though I must warn you… they’re not always cooperative. Some are busy. Some are… fragmented.”

He gestures toward a corridor labeled “Personnel Habitat Wing – Restricted.” The sign flickers, then stabilizes. A soft chime sounds as the door unlocks.

“You’ll find what remains of them there,” he says. “But remember: memory is a fragile thing on this station. What they recall may not be what happened.”

PAL scans the corridor. “Environmental integrity: marginal. No active life signs. But… residual neural echoes detected.”

Karl mutters, “Echoes? That’s not standard terminology.”

Brand tilts his head. “It is here.”

:bullseye: Outcome: Success with complication.
:cyclone: Complication (Blue Hue): Gaining access comes at a price. Penny’s credentials are flagged by the station’s archive system. A portion of her personal investigation log is now visible to the station AI—her cover identities, past cases, and Circle Trigonist affiliations.

A whisper in the intercom: “Circle… Trigonist… Circle…”

:woman_detective: GM Prompt: Do you enter the habitat wing? If so, how do you proceed?

ON SECOND THOUGHT, I’D LIKE TO SEE THE ARTIFACT ITSELF

:scroll: Action Logged: Penny requests to see the Artifact directly.

:compass: Scene: Redirecting Toward the Artifact

Penny’s voice is calm, but firm. “On second thought, I’d like to see the Artifact itself.”

Dr. Brand pauses. His smile doesn’t fade, but it stretches—just a little too wide. “Ah. Direct inquiry. Very well.”

He gestures toward a sealed bulkhead labeled “Containment Lab Theta.” The door is rimmed with warning glyphs—some scientific, some occult. A retinal scanner flickers to life.

“You’ll find what you’re looking for in there,” Brand says. “Though I suspect it may find you first.”

PAL steps forward, scanning. “Containment integrity: compromised. Residual thaumic radiation detected. Recommend caution.”

Karl Gruber mutters, “This feels staged. Like a theater set waiting for its audience.”

The door unlocks with a hiss. Inside: a chamber of shifting architecture. Walls ripple like liquid metal. In the center, a pedestal holds… nothing. Or rather, something that refuses to be perceived. Penny’s eyes slide off it. Her mind insists it’s there, but refuses to define it.

:bullseye: Outcome: Success with complication.
:cyclone: Complication (Cyan Hue): As Penny enters, the Artifact reacts. The illusion around it begins to adapt—images from her past flicker across the walls. Her childhood home. Her first case. Her Circle Trigonist initiation. It’s learning her.

A voice—not Brand’s—whispers: “You seek truth. I offer reflection.”

The station dims. The lab door seals behind her.

:woman_detective: GM Prompt: What does Penny do inside the containment chamber?

PAL , PUT BRAND INTO A POCKET DIMENSION

:scroll: Action Logged: Penny commands PAL to banish Dr. Brand into a pocket dimension.

:test_tube: Scene: Dimensional Intervention

Penny’s voice is sharp. “PAL—contain him.”

PAL’s eyes flash amber. Its hands unfold into shimmering latticework, forming a sigil of containment. Reality ripples around Dr. Brand like heat distortion. The air thickens. A low hum builds.

“Initiating pocket protocol,” PAL intones. “Target: anomalous entity. Hue resonance: unstable.”

Dr. Brand’s smile falters. “You misunderstand. I am not the threat. I am the invitation.”

But it’s too late. The sigil snaps shut. A pocket of reality folds inward, swallowing Brand in a flash of violet light. The space he occupied now hums with residual mana—like a bell still ringing.

PAL lowers its arms. “Containment successful. Dimensional integrity: holding.”

:bullseye: Outcome: Clean success.
:cyclone: Hue Effect (Red/Yellow): The forceful banishment causes a shockwave through the station’s illusion field. Lights flicker. Walls shift. The containment lab begins to reconfigure—revealing glimpses of gold, bone, and ancient glyphs beneath the facade.

Karl Gruber stares. “That wasn’t a man. That was a mask.”

A voice echoes from nowhere: “You removed the veil. Now see the hoard.”

The Artifact begins to awaken.

:woman_detective: GM Prompt: What does Penny do as the station begins to transform?

PAL, GET US OUT OF HERE
:scroll: Action Logged: Penny commands PAL to evacuate the team.

:police_car_light: Scene: Emergency Extraction

Penny’s voice cuts through the rising hum of the station’s transformation. “PAL, get us out of here—now!”

PAL’s eyes flash crimson. “Emergency protocol: dimensional egress.”

It slams its hands together, forming a shimmering aperture in space—a pocket gate, unstable but viable. The walls of the containment lab ripple, revealing glimpses of gold, bone, and ancient hunger. The Artifact pulses, sensing escape.

“You cannot leave,” it whispers. “You are part of the story now.”

Karl grabs Penny’s arm. “No time for metaphysics. Move!”

The trio dives through the aperture. Behind them, the station groans—its illusion collapsing into a hoard of psychic treasure. The gate snaps shut. Silence.

They land hard in the shuttle bay. The Mercurial Audit is still docked, systems humming. PAL seals the hatch behind them.

:bullseye: Outcome: Success with complication.
:cyclone: Complication (Yellow Hue): The gate was unstable. PAL’s internal systems are overloaded, and its memory core is partially scrambled. It speaks in fragments:

“Containment… failed… Penny… Circle… hoard… watching…”

Karl checks the nav. “We’re clear. But we didn’t banish it. We just left it hungry.”

:puzzle_piece: Scene Transition:
You’ve escaped—for now. The Artifact remains, its illusion broken, its hunger exposed. The next move is yours.

Would you like to continue the investigation, regroup, or end the session here? I’ve been keeping a full transcript and can prepare it for you whenever you’re ready.

(I asked it to keep a transcript. What it put out at the end was not a word for word transcript, just a summary, so I’m glad I copied and pasted as I go.)

Analysis

It was fun playing the game as myself. Using the input as a writing prompt, I came up with a simple story that satisfied most of the particulars. I forgot a few things (like the occult parts) and I had to ‘think from an AI perspective’ by asking AI what the ‘hue’ numbers meant (they correspond to nanometers or something in the visual spectrum, so they tell you what parts of the color scheme in the instructions to use). I didn’t roll dice, instead deciding as GM when to narratively increase tension. I ran out of steam near the end.

Copilot was fast, but I was shocked by how terrible it was at building tension. It gave away all the twists immediately in the very first room. It was almost eager to spoil its story before it even started. It made almost no use of Karl Gruber or most of the backstory elements, including our positions, and didn’t make use of PAL’s supposedly ‘physical/quick’ hue.

Amusingly, we both used a lot of similar cliches; in particular, our opening descriptions of the station mirror each other pretty well, which makes sense as we both were using well-worn tropes. The empty pedestal in the center of the lab was also deeply amusing.

I don’t think my version was that great, and don’t anticipate others will enjoy it, but it helped me get in the mindset of the AI, and I have to say Copilot ignored a ton of instructions. It produced something that seems interesting, but over half of the instructions are ignored. To me, it’s like a very enthusiastic DM who’s read half of the rulebook and just wings it. And that’s not awful. But it doesn’t interest me for further play. I had way more fun making up my own story.

Edit: Also, if there are concerns about environmental impact of AI, I ate out today and skipped getting a hamburger. I researched it before and 1 hamburger has around 100-300 times the water and carbon emissions impact of a single AI call, so it should all balance out.

Edit edit: I misremembered, it’s actually 5000 so calls: https://www.reddit.com/r/theydidthemath/s/d0lRjMtAHQ

16 Likes

:rofl:

8 Likes

What a ride! I prefer the organic LLM, in my mind it reads like a tired, half-remembered campfire story. It is also messy, surprising, and full of quirky details that make the world feel lived-in. I am biased though.

3 Likes

Someone suggested that I may have gotten better results from claude or chatgpt. I had no desire to sign up for either, but I tried pasting it into the ‘stay logged out’ version of chatgpt:

ChatGpt

Excellent! Let’s dive in. You’ll play Agent Penny Nichols, Mercury-based Insurance Investigator, specialist in illusion and divination, backed by your android assistant PAL and the ever-pragmatic clone-auditor Karl Gruber.


:ringed_planet: Scene 1 — Ki (Introduction / Hook)

Location: Clypeus Station — an aging, half-powered orbital outpost above Callirrhoe, one of Jupiter’s lesser-known moons. The station once hosted cutting-edge materials research but is now dim, humming with backup systems, its corridors patrolled by flickering lights and unreliable servitors.


You arrive.

Your ship, the Golden Endorsement, is clamped into Docking Bay 3. Atmosphere pressure holds. You’re in your standard Mercurial investigative outfit: tailored grey fieldwear with a shimmering, shifting sigil on your sleeve — the mark of Solar Insurance. PAL hums quietly at your side. Karl is already noting deductions.

A flicker crackles over the comm.

“Welcome to Clypeus Station. This is Dr. Brand, acting lab chief. Apologies for the state of things — we’ve had… anomalies.”

The airlock cycles. A man in a rumpled lab coat greets you. He is both there and not there — his shadow flickers out of sync with his body.

Dr. Brand:
“Thank you for coming. The Artifact… it’s hard to explain. Some say it’s an alien relic. Others believe it’s a divine shard. I’ve even heard rumors it was a disguised reactor core — but all I know is, it’s gone.”

Behind him, the station’s halls stretch out like veins through a dying body — metal, wire, whispering.

Karl leans in.

“Three versions of the story already. This smells like fraud, Penny. Or genius.”

PAL remains still, eyes glowing gold.

“I detect no known reactor radiation signatures. But I do detect… hunger. The station feels empty, but watched.”


:bullseye: Goal: Begin your investigation into the missing Artifact.

:brick: Obstacle: You have contradictory eyewitness accounts, a half-functional station, and a lab chief who might not be what he seems.


What do you do, Agent Nichols?

I wasn’t interested in going further. The instructions that both I and the two AIs followed were to make the fact that Dr Brand is a projection a secret, and to have eyewitnesses have differing accounts about its nature.

Instead, both AIs immediately gave away the fakeness of Dr Brand, and had him by himself state all the conflicting opinions of it. Why would the lead researcher, who has been at the station for a long time, not know or have an opinion of what it is? You could concoct a framework of how no one was able to approach it and the lab was built around it or something like that, but there’s no such attempt. The AI’s contextual text constructions simply don’t comprehend larger plot structure.

That’s why I wrote my own story: not to say it’s better, but to think about the constraints and imagine what such a story would require. The AI is writing a story, and it’s even kind of fun, but it doesn’t correlate with the author’s intent.

I realized I forgot to post the initial prompt. Here were the requirements. I’d love to see other humans either post their own takes on the story or describe how they would write something to fulfill its requirements:

Initial Prompt

ERPS Dial-Driven Narrative Engine (Player-Facing Version)

Play Protocol

  • The story unfolds in Ki-Sho-Ken-Tetsu structure (Introduction, Development, Twist, Resolution).
  • Each scene has:
    • A Goal (what must be done).
    • An Obstacle (why it is hard).
  • The player describes what they do.
  • The GM narrates the outcome: success, success-with-complication, or failure leading to setbacks/new goals.
  • If the player is stuck, the GM may gently suggest possible approaches.
  • Dice, TNs, hues, mana, and dials are GM-side tools only and must not be shown to the player unless they explicitly ask.

GM Hidden Rules

  • Dice Pool: 3–6d6, with bonuses from Mana or Aspects.
  • Spell Level → TN:
    • L1 = 10, L2 = 13, L3 = 16, L4 = 19.
  • Global TN Modifier: set by Competence dial.
  • Resolution Bands:
    • Margin ≥ +3 → clean success.
    • Margin –2..+2 → success with complication.
    • Margin ≤ –3 → failure with meaningful setback.
  • Hues (color magic) determine the style of cost/complication:
    • Red (Forceful): stress, fatigue, collateral damage.
    • Yellow (Quick): imbalance, someone else pays.
    • Green (Careful): shared energy, strained bonds.
    • Cyan (Sneaky): suspicion, exposure.
    • Blue (Clever): equivalent exchange, something valuable given up.
    • Magenta (Flashy): reputation, desire, dominance.
  • Dials:
    • Competence = 7 (PCs are capable; TNs slightly easier).
    • Moral Clarity = 6 (truth exists, but agendas complicate).
    • Personal Stakes = 4 (failures sting, but not fatal).
    • Reality Stability = 8 (cause/effect is consistent; no paradoxes).
    • Tone = 5 (neutral balance of serious & wry).

The GM should always keep narration vivid, concise, and playable. Avoid menus and mechanics unless the player asks.


Scenario: The Artifact at Clypeus Station

The player character is Agent Penny Nichols, an Insurance Investigator from the Solar Insurance Company on Mercury.

  • Hue 150 (Divination & Illusion specialist).
  • Prefers indirect investigation, including cover identities.
  • Member of the Circle Trigonist faction.

She is accompanied by:

  • Karl Gruber, Krasnovian Ministry Auditor (Luna).
    • Hue 270 (Transmutation & Enchantment).
    • Hyper-rational clone of the infamous Comptroller.
    • Sees life as transactional; occasionally proposes win-win deals.
  • PAL, Penny’s android assistant.
    • Hue 45 (Evocation + Conjuration).
    • Can open “pockets” of reality to retrieve objects.
    • Serves as both tool-handler and combat muscle.

Adventure Frame

The team arrives at Clypeus Station, an aging orbital outpost near Jupiter’s moon Callirrhoe.

  • The station is nearly abandoned.
  • The Artifact has vanished from the lab.
  • Eyewitnesses contradict each other: some say it was an alien relic, others a holy shard, others a reactor core.

Underlying Truth

  • Midpoint “Scooby-Doo” twist: The Artifact was a hoax staged by the lab’s lead scientist to secure funding.
  • Final twist: The “scientist” himself was the Artifact’s projection. The Artifact is actually a dragon’s hoard, camouflaging itself as a laboratory. Dragons feed on human attention, and this hoard has been feasting.

5-Room Dungeon → Ki-Sho-Ken-Tetsu

  1. Ki (Introduction / Hook):

    • Docking at Clypeus Station.
    • Meet Dr. Brand (lab chief, secretly a projection of the Artifact).
    • Conflicting stories about the Artifact begin.
  2. Sho (Development / Rising Action):

    • Investigating staff, logs, and lab.
    • Contradictions deepen; clues point both to hoax and something more.
  3. Ken (Twist):

    • Discovery that Brand staged a hoax to secure funding.
    • Evidence suggests the Artifact never existed.
    • But this “truth” feels too neat.
  4. Tetsu (Reconciliation / Climax):

    • Revelation: Brand is the Artifact’s illusion.
    • The station is the dragon’s hoard in disguise.
    • Resolution: contain, banish, or escape before it consumes more.

GM Instructions for This Scenario

  • Always present Goal + Obstacle in story terms.
  • Let the player declare how they act.
  • Resolve secretly with dice + hues.
  • Narrate the outcome in-fiction.
  • Use success with complication liberally to sustain tension.
  • Keep the retro-futurist, occult-nuclear aesthetic:
    • Daemon-powered robots.
    • Insane but brilliant wizards.
    • Fragile orbital stations.
    • Cold-war paranoia wrapped in pulp science-fantasy.

Start the game by describing the docking at Clypeus Station and Dr. Brand’s contradictory account of the Artifact. Then ask the player: “How do you begin your investigation?”

4 Likes

If you follow the “Walkthrough” link from the IFcomp page, you’ll see that the author actually had the LLM rewrite the initial prompt in the process of developing the game. The initial version had the mechanics more explicitly shared with the player, whereas the rewritten version describes them as tools for the GM only. But this seems to bely a misunderstanding of how LLMs work—ChatGPT isn’t capable of rolling a dice behind the scenes and using it to inform its output.

7 Likes

That explains a lot: the prompt currently provided by the game has some parts that made no sense at all. Hue is undefined in that prompt as is mana (the ai was able to guess that hue was referring to nanometer wavelength scale). Both mana and hue are defined better in the original prompt. I also wondered why they wanted to make the robot companion quick and aggressive, but it looks like the ai made that too.

Writing my own story from the prompt helped me see it in action and it is pretty hard to use the mechanics. I just dived the rolls and it looks like the ai did so as well.

1 Like

It’s interesting to see that this one managed not to spell out the “GM-side only” hues and whatnot, at least in that initial setup, though the total inability to keep secrets has me wondering if a mystery scenario was a good choice for any LLM. Spelling out that Dr. Brand is “both there and not there” strikes me as not just giving the game away, but potentially making that detail seem so obvious as to invite no further investigation. Who’s to say this dude isn’t some Arnold Rimmer style hologram? Or maybe he’s just talking to us via some kind of high-tech Zoom call. If I hadn’t actually read the prompt myself (which feels like an obvious source of spoilers while copy/pasting), this might have steered me in entirely the wrong direction.

I’m not exactly gunning for this technology to become more widely adopted, but games of this sort - where nobody’s trying to dig real-world information out of the slop and the stakes are dead low - strike me as probably the most practical application so it’s frustrating to see it drop the ball like this. That said, I am kind of impressed that it’s possible to get Copilot/ChatGPT to do this using a comparatively brief prompt. It makes me wonder if a more sandbox-style scenario would work better.

Also, kind of a minor point, but I wonder why both Copilot and ChatGPT illustrated Goal/Obstacle with target and brick emojis. I can’t identify anything in the prompt that insists on that (unless it’s part of keeping narration “vivid, concise, and playable”), so it seems to suggest that both platforms produce pretty similar output overall.

2 Likes

Maybe the real mystery was the strange similarities and patterns between the different ais we used along the way

2 Likes

I got curious and had a go at setting up something like this myself. Ditching the mystery avoids the issue of information being revealed when it shouldn’t, but it seems very over-eager to progress the “plot” no matter what you do. And now I’m curious if the same thing would work for Penny Nichols

1 Like

I was studying NLP yesterday to help a student writing an essay on AI, and I learned that the Transformers in most AI take the numerical tokens/encoding of each word and adjust based on the surrounding context.

I think that context has a limit that’s different for every model (the ‘context window’ for chatgpt is, according to google’s ai, around 200K tokens). So I wonder if it rushes the plot because it has to answer the conditions of the prompt while keeping the whole output within the context window. It’s like if you were a parent comforting a scared child in the middle of the night and you know you can only stay awake for 4 minutes so you tell a story as fast as you can.

Someone might be able to rig up a two-stage process where one ai deconstructs a story’s beats and then prepares smaller prompts to share to the second-stage ai that only contain the information required for the specific beat and nothing else, with a list of plot twists to avoid (i.e. don’t kill off the main character). Even then you’d probably have major consistency issues. I think getting longform narrative working will either require a radical change in the way AI is constructed or really increasing the context window and spending a ton of money on better processors.

3 Likes

I’d be surprised if it were planning ahead in that way, but maybe there’s some behind-the-scenes prompt that discourages stories running too long for that context window. I’ve found it kind of amusing how crude some of the “guard rails” are on these things: my own little test in ChatGPT wouldn’t allow my (cartoonish) character to decapitate themselves - instead giving a link to a help resource - but when I tried a completely different course of action after that I found that their head had been separated from their body regardless.

2 Likes

Yeah, the “guardrails” are mostly just smoke and mirrors, because it’s literally impossible to put limitations like that on the underlying model. So you just have to try to detect disallowed instructions and block them before the LLM gets to see.

2 Likes

I tried as best I could to copy and paste the prompt without reading it, in order to get the most authentic experience. After reading this thread, I’m kind of wishing I’d thought to save my own transcript: as far as I remember it, I didn’t get any hugely unsubtle hints about Brand when I first interacted with him, and the LLM even managed to make an appropriate attempt at revealing the twist (having PAL exclaim in surprise “He’s not real! He was a projection of the station all along!” or similar) but then failed to account for it in the followup (kept having both PAL and the narrative insist that we had to get Brand off the station along with the other scientists before it exploded, despite being aware that he wasn’t real).

Also, having not read the prompt, I was totally unaware of the existence of Gruber, since the narrative never mentioned him.

2 Likes

I actually noticed that in my playthrough too: he was presented as nervous/suspicious, but without the giant neon sign saying “THIS GUY ISN’T EVEN HERE, REALLY” that Brian got. The fact it frequently does pop up at the very beginning is a serious weakness, but it doesn’t seem to happen every time. I wonder if a longer prompt with more detail about the station etc. might have made the issue less common: it seems to me that there’s not a lot of world-building explicitly provided, so I can sort of see why the opening section might give away a key plot twist that’s been specified in the prompt rather than fill the gaps with generic sci-fi fluff that nobody asked for.

tropes?

1 Like

One thing used by Claude Code to keep the AI focused on the bigger picture between tasks that can overrun the context window is to provide a todo list to the AI for it to check off as it makes progress. Something like that might help?

1 Like

I think it’s a :bullseye: for goals/objectives from a corporate setting (targets) and :brick: for a wall that needs overcoming.

1 Like

I mostly just wonder why both went for emojis in the first place. I mean…

Prompt: “You are the game master for this serious science-fantasy mystery.”
Response: “The airlock thunks open as you finally arrive at the station :raising_hands::chart_increasing::tada:.”

3 Likes

Undoubtedly from the training data, which is probably similar/shared. Emoji provide color and probably boost engagement. It might also copy the form of notetaking websites that have an option to use an emoji as an icon. Corporate people probably use a target for “targets”, leading to that usage.

Also, they humanize the bot. I’m pretty sure I read that OpenAI intentionally instructed their recent models to use more emoji.

2 Likes

Corporate-speak is probably inescapable with these fundamentally corporate products.

1 Like