Has anyone here tried using Chat GPT or other AI for solo play ? I have just spent a few hours feeding details from Double Adventure 1 Shadows into Chat GPT - maps, location descriptions etc. and then playing off the back of that. No combat yet, just lots of rolls, searching etc It needs more time and work but I feel that this has a real possibility of leading to the scenario being able to be played totally solo.
Chat GPT for solo
- Thread starter Saraband
- Start date
I have searched on this forum and not found anything.
Yes. Chat GPT already knows the rules and with a little prompting it will start playing a scratch game with it acting as GM, it can generate a character for you or you give it one, or you can play manually through character creation with it. It then will set up a simple game. ie "You're at the startport and you see a fight happen, what do you do ?" (gives 4 options).
I then asked if we could play "Shadows", it said "yes, I know Shadows", and it started setting up a situation like "Shadows" but different eg different planet (not Yorbund). I said "No Shadows is on Yorbund". It said "Oh you want to play canonical Shadows". It then started a game which was slightly more like "Shadows" but not "Shadows". So then I said I started pasting in slabs of the book (you have to edit slightly). It took all that on board and used it. For example I pasted in a map and it understood the map ie you are location 22 - you can exit to location 21 etc. Then I pasted in descriptions with die rolls for discoveries and it incorporated that.
I now need to get the whole book and slightly edit it and feed it in. I am pretty confident that this is workable. Quite time consuming but workable.
I have done Traveller Solo with ChatGPT several times.
I posted the transcripts on the Cleon Memorial Library subforum here on the board.
I think it was CyborgPrime that set up a Traveller game in Chat GPT as well.
ChatGPT seems to work really well for this. Almost as if it read every rulebook.
I posted the transcripts on the Cleon Memorial Library subforum here on the board.
I think it was CyborgPrime that set up a Traveller game in Chat GPT as well.
ChatGPT seems to work really well for this. Almost as if it read every rulebook.
Not solo play (yet) but I've been playing around with ChatGPT a little ... and spent some time seeing if it can be useful as a referee's assistant between sessions. I fed it source material about the 5FW and asked it some questions. Two shortcomings so far:
Also, while it claims to know multiple editions of the rules, when I ask it to use T5 it agrees and then doesn't. (I'm not sure if it's defaulting to CT or MgT or something similar.)
But all of this could be because I'm still new to AI. A better prompt might work.
On the flip side, if the players walk into a bar and you need a quick menu appropriate to the local, ChatGPT is better at generating such things in an instant. Food, drinks, clothing, music styles ... all good. So, it's useful in that regard.
- If a fleet is at one location on one date, and at another on another date, the AI doesn't understand (so far) the implication that between those dates the fleet is somewhere ... the original location, the final location, or an intermediate location.
- It struggles with understanding the distance between two hexes. After an hour of repeated attempts with me correcting and re-explaining each time, it's still usually off, but now only by a parsec each time.
Also, while it claims to know multiple editions of the rules, when I ask it to use T5 it agrees and then doesn't. (I'm not sure if it's defaulting to CT or MgT or something similar.)
But all of this could be because I'm still new to AI. A better prompt might work.
On the flip side, if the players walk into a bar and you need a quick menu appropriate to the local, ChatGPT is better at generating such things in an instant. Food, drinks, clothing, music styles ... all good. So, it's useful in that regard.
"Also, if you're on the free tier at OpenAI, there's a three documents per day limit." Yes I've hit that limit.
You're totally right about the minor details - food, drink etc Chat GPT is great at generating that. And is pretty good at generating abstract Traveller-esque gameplay.
As I say above I was trying to achieve something a bit more structured - in the case above running a Trav adventure via Chat GPT. I was amazed that when I pasted in maps and put myself in say Room 1 and then asked it what were my options - it accurately gave me the exits to the other rooms (ie "there is an exit to the south, the west" etc). Having said that it all starts to break down without more directed prompting
Now I am trying to re-write the entire adventure with prompts eg feeding in room descriptions and then dividing the descriptions into "Player's knowledge" and "Referees (ie Chat GPT) knowledge". My goal is to generate word document that I can give to someone else and by just feeding in the prompts without necessarily reading them they can have a full solo adventure.
Just as an example from Shadows with a few prompts I got the entry to the pyramids to this, -
Now you have the possible entry points based on what Joren and Harlan can investigate. What will they do next?
You're totally right about the minor details - food, drink etc Chat GPT is great at generating that. And is pretty good at generating abstract Traveller-esque gameplay.
As I say above I was trying to achieve something a bit more structured - in the case above running a Trav adventure via Chat GPT. I was amazed that when I pasted in maps and put myself in say Room 1 and then asked it what were my options - it accurately gave me the exits to the other rooms (ie "there is an exit to the south, the west" etc). Having said that it all starts to break down without more directed prompting
Now I am trying to re-write the entire adventure with prompts eg feeding in room descriptions and then dividing the descriptions into "Player's knowledge" and "Referees (ie Chat GPT) knowledge". My goal is to generate word document that I can give to someone else and by just feeding in the prompts without necessarily reading them they can have a full solo adventure.
Just as an example from Shadows with a few prompts I got the entry to the pyramids to this, -
Entry Point 1
- Description: A shaft on the platform with vents or drainage pipes at its base. The shaft is about 10 meters deep.
Entry Point 2
- Description: A statuary figure on the platform or atop the pyramid may be movable to reveal an entry.
Entry Point 3
- Description: There may be a concealed doorway in either of the stairways leading up to the pyramid.
Entry Point 4
- Description: A concealed shaft entrance may be hidden on the face of the pyramid, probably in shadow.
Entry Point 5
- Description: The flat upper surface of the small pyramid might have an entry.
Entry Point 6
- Description: Brute force can be applied to break through the pyramid’s surface using an energy weapon.
Now you have the possible entry points based on what Joren and Harlan can investigate. What will they do next?
100%, I now have seen your thread s on this website. Good one
"Note that if you feed it documents, they don't persist beyond the current chat (though you can revisit the same chat again and again over multiple days). Also, if you're on the free tier at OpenAI, there's a three documents per day limit."
I'm not sure this is a problem. I am able to access old threads (ie weeks ago) on Chat GPT, and I just have the free service. It seems to remember. Just as a side note - Chat GPT tells me that it doesn't use my Chat GPT information to interact with other people. I'm not sure that that is strictly true otherwise how is it getting so good at running Traveller ? Anyway I wish it did then it could learn from other people and use that with me.
Cardboard Heroes Traveller
In Settings --> Data controls, there's a "Improve the model for everyone" switch.
In Settings --> Data controls, there's a "Improve the model for everyone" switch.
If you want to take a look at my Prompt, I posted it the other day. I'd love to see how you change it.
EDIT: You can also ask CyborgPrime about his Virtual GM. It was pretty cool.
Last edited:
If you want to take a look at my Prompt, I posted it the other day. I'd love to see how you change it.
EDIT: You can also ask CyborgPrime about his Virtual GM. It was pretty cool.
I had a look at your prompt cheers.
This is what I start with (the stuff about not telling me the details of the rolls, was related to me previously pasting in a lot of room details, some of which had random events, which would come to players as a surprise - in this case mainly animal attacks) -
ME - do you remember how i like to play traveller ?
ChatGPT said:
Yes, I do!You like to play Classic Traveller (CT) solo, and you prefer a rules-heavy style that sticks closely to the published rules. You want full transparency on things like dice rolls and decision points, and you want time tracked carefully when you're exploring.
When a dice roll is needed, you like me to tell you just the details of the roll (not the consequences yet), and then only explain the consequences if the roll succeeds. Also, you prefer that I don't reveal referee knowledge unless you specifically ask for it.
Nice!
I'll try to incorporate that into my prompts.
I'll try to incorporate that into my prompts.
Cheers Spinward good luck with your gaming.
For anyone reading this, this thread from my Chat GPT gives a good idea of its strengths and limitations (G and J are my two characters) -
CHAT GPT -
Just across a narrow alley from Portside Eats, there's a squat, prefab structure with flickering red signage reading "Sleep-Eeze Pods."
It's a low-end capsule hotel, built like a stack of shipping containers welded into the side of a service block. Each unit is just big enough for one person to crawl into, featuring:
The front desk is just a reinforced kiosk with a bored attendant behind a scratched plaz window. A sign reads:
ME -
"we want two pods for the night, but do you have anything cheaper ?"
CHAT GPT -
The clerk at the low-end accommodation block — a wiry woman with a creased synth-coat and a dull wristcomp — gives G a once-over and shrugs.
Each tag is marked in faded red stencil: POD 42 and POD 43.
The pods await. Do G or J do anything else before turning in?
She pauses mid-motion, narrows her eyes, and jerks her chin toward a cracked sign above the counter:
Want G or J to say or do anything?
For anyone reading this, this thread from my Chat GPT gives a good idea of its strengths and limitations (G and J are my two characters) -
CHAT GPT -
Just across a narrow alley from Portside Eats, there's a squat, prefab structure with flickering red signage reading "Sleep-Eeze Pods."
It's a low-end capsule hotel, built like a stack of shipping containers welded into the side of a service block. Each unit is just big enough for one person to crawl into, featuring:
- A filtered air feed
- Fold-down bunk
- Basic privacy screen
- Shared shower and sanitation unit
- Coin-op lockers bolted to the wall inside a common area
The front desk is just a reinforced kiosk with a bored attendant behind a scratched plaz window. A sign reads:
ME -
"we want two pods for the night, but do you have anything cheaper ?"
CHAT GPT -
The clerk at the low-end accommodation block — a wiry woman with a creased synth-coat and a dull wristcomp — gives G a once-over and shrugs.
She taps a worn display with a cracked corner. Faint red digits scroll past.
What do GJ decide?
You said:
J says "no we'll take two 15 cr pods"ChatGPT said:
The clerk nods, barely looking up.She slides two thumb-sized coded tags across the scratched counter.
Each tag is marked in faded red stencil: POD 42 and POD 43.
The air smells faintly of disinfectant and cheap polymer. The sound of distant machinery hums through the thin walls.
The pods await. Do G or J do anything else before turning in?
You said:
does she see J's rifle ?ChatGPT said:
Yes—J’s rifle is slung over his shoulder. The clerk clocks it immediately.She pauses mid-motion, narrows her eyes, and jerks her chin toward a cracked sign above the counter:
She eyes J a moment longer, then shrugs.
No alarm raised—for now. Just a warning.
Want G or J to say or do anything?
I'm possibly raving to myself here, but Chat GPT is really at a high level.
What I did was feed the rumour table from Adventure 1 Kinunir into Chat GPT, then it does a weekly roll and generates a rumour letter (ie A, B, C, D etc), it then retrieves the rumour. It does that well and easily. Then without showing me the rumour, it then (after me telling it to) contextualises the rumour.
Here is the rumour as it appears in Adventure 1 Kinunir - "A naval veteran (age 43, 246869) served on the Allamu right up to its last flight, when he stayed behind suffering from fever. Through all his years, he has carried his access shaft maintenance key as his good luck charm. He can describe the access shaft on the Kinunir class of ship, and demonstrate how the key would work. -He does not want to part with the key itself."
Here is the Chat GPT text -
Standing by to present the content of Rumor E, Chat.
Waiting, Chat — no action until you say so.
Yes, Chat can do that exactly:
When a rumor result like "E" is rolled, I will:
Fully understood and ready to proceed when you say.
Rumor E — embedded in context:
It’s late afternoon at Portside Eats, and the usual hum of worn-down conversations buzzes through the diner’s filtered air. G sits nursing a bitter cup of synth-caff, while J flips through a static-filled local news feed on the wall unit.
A man in his 40s at the next table leans over — his voice gravelly, his breath laced with cheap alcohol. He wears an old Navy surplus jacket, its shoulder patch half-torn but still readable: Allamu.
Just incredible. And look at what it withheld - it didn't tell me his UPP, because of course a real ref wouldn't tell me that.
What I did was feed the rumour table from Adventure 1 Kinunir into Chat GPT, then it does a weekly roll and generates a rumour letter (ie A, B, C, D etc), it then retrieves the rumour. It does that well and easily. Then without showing me the rumour, it then (after me telling it to) contextualises the rumour.
Here is the rumour as it appears in Adventure 1 Kinunir - "A naval veteran (age 43, 246869) served on the Allamu right up to its last flight, when he stayed behind suffering from fever. Through all his years, he has carried his access shaft maintenance key as his good luck charm. He can describe the access shaft on the Kinunir class of ship, and demonstrate how the key would work. -He does not want to part with the key itself."
Here is the Chat GPT text -
ChatGPT said:
Kinunir Rumour Table roll: Row 1, Column 5 → Result: EStanding by to present the content of Rumor E, Chat.
You said:
waitChatGPT said:
Waiting, Chat — no action until you say so.You said:
so what I want you to do is get the text of the rumour and present it in a context of a place and a conversation, can you do that ?ChatGPT said:
Yes, Chat can do that exactly:When a rumor result like "E" is rolled, I will:
- Retrieve the full text of the corresponding Kinunir rumor (which you've already pasted).
- Embed it naturally into a scene — placing it in a believable location, through a specific person, in a grounded conversation that fits CT tone and your setting.
Fully understood and ready to proceed when you say.You said:
bang !ChatGPT said:
Rumor E — embedded in context:It’s late afternoon at Portside Eats, and the usual hum of worn-down conversations buzzes through the diner’s filtered air. G sits nursing a bitter cup of synth-caff, while J flips through a static-filled local news feed on the wall unit.
A man in his 40s at the next table leans over — his voice gravelly, his breath laced with cheap alcohol. He wears an old Navy surplus jacket, its shoulder patch half-torn but still readable: Allamu.
He holds it out for a second, then pulls it back.
He takes a sip from his cracked glass and leans back, already fading into the low noise of the diner again.
Just incredible. And look at what it withheld - it didn't tell me his UPP, because of course a real ref wouldn't tell me that.
