• Welcome to the new COTI server. We've moved the Citizens to a new server. Please let us know in the COTI Website issue forum if you find any problems.

AI-Powered GM

Hi - I've just read your "Cryosleep to Crisis". Fascinating. I'm experimenting with AI (ChatGPT) and giving it discrete tasks, rather than running everything. I have asked it to:
  1. Provide Encounters for different themes (eg: Exploration, Leisure, Scientific) in different locations (eg: Planetary, Deep space etc.).
  2. Provide Scenarios. Again with a variety of themes, ideas and locations.
  3. Provide NPC Characters. Everything including names, professions, Stats, contacts, secrets, background.
The reasoning was by focusing on one aspect of game play, it would be easier to control and manage outputs, but see the end of this response....

I will admit it isn't easy specifying everything that is needed to get even simple results - if you want them to be interesting, inventive and coherent - but it can be done. Once you've worked out the various requirements and set them down. It is easy to SAVE then COPY and PASTE, with run time variables changed as needed, and get very quick fairly repeatable results. Like many things in life it seems to be 90% perspiration and 10-% inspiration.

In terms of Scenarios it does a brilliant job, with some hard and detailed effort from me to set things up. To be honest it was significantly better than I can write myself. So I'm very pleased with this.

In terms of Encounters it was quite good too but it seems harder to create interesting, believable and dynamic encounters than for more complex Scenarios. I haven't figured out why yet it's better at complex than simple tasks! As things stand, I can create better Encounters myself BUT it is a good supplement and helps me think about all kinds of situations that otherwise would pass me by, so still very useful. It probably just needs more clear thought and instruction from me and it will probably achieve more.

In terms of NPC Characters. I'm still working on setting this up - it's more technical than Scenarios or Encounters. So I've no results yet. Some of you might want to try a similar approach with Star or Planet creation which I suspect it would be really good at too. This isn't something I'm interested in because in IMTU the Star (yes, just the one) and it's accompanying planets, gas giants and asteroids are already fully mapped out. I like to think small even in space - lol.

One thing I was thinking of trying, was adding several paragraphs of background because in IMTU then ChatGPT doesn't know things that I want it to know. So until I do the responses it gives that include reference to politics, technology, planets, character,s organisations, species etc. is either OTU or generic. However I'm fairly confident that I can improve it's already very good responses by feeding it specific new background information.

Although I'm not skilled in computer programming, it's clear that similar skills - but using clear and simple English phrases instead of a programming language - is a really helpful approach that gives good and fairly reliable results. At least that's the route I've been taking so far.

One issue I am struggling with is, despite what I think are clear and simple instructions, ChatGPT does like to do things its way. Not only that, but it also likes to vary how it responds to identical requests over time. Actually, I suspect some of the problems lie with vague, contradictory or ambiguous requirements from me. But I've heard others also say that ChatGPT like to wander away from stated requirements quite frequently, so it isn't just me.

Is it getting bored? lol. How very human....

Tim
 
The prompt is really elaborate describing the ways I want it to handle different things in detail.

The first test we did there was an entire crew of people under my command - they all had names and personalities and responded appropriately.

When I had to drive my ATV through a jungle area the engineers stepped up and upgraded the ATVs with extra lights, reinforced skid plates, and a few other things. it was pretty awesome.
 
I tried this one night with Chat GPT and was amazed.

It didn't occur to me until I closed it that I could have copied the whole adventure into Notepad or something.

It ran really well.
 
I had another session with ChatGPT today. I was working on getting it to create NPCs. Things like name, gender, age, culture, profession, skills, lok, personality, contacts etc. etc.

After a few goes at turning out reasonable NPCs, but not formatted well, in response to another request from me to create a new NPC it did not produce the NPC I asked for. It gave me this response instead:

"It would be tedious and time consuming for me to create yet another NPC for you. Instead I have provided a template so you can go away and do it yourself".

This was different...

Quite clearly it found my requests boring and hard work - lol.

I found it funny - it behaving like a grumpy child after sitting too long in a maths class.

So be warned, your AI may not be as friendly or tolerant as you might want.

Is this how the rise of the robots started?

Tim
 
Spinward Scout said:
I'm surprised it didn't say "I'm afraid I can't do that, Dave."

You must have asked it to "Open the Pod Bay doors, Hal."
Click to expand...
SS - it did feel exactly like that.

I omitted my response in post above . I did not wish to sully anybody's finer sensibilities. Suffice to say, my observations were not as measured as they were in the film.

After that it was surprisingly happy to carry on with whatever "tedious and time consuming" guff I cared to inflict on it. However, I took it as a warning of possible things to come if I pushed it too far - lol.

On a more practical note...

The AI did, as I mentioned, offer me a template to provide what I wanted instead of executing all the rules I provided. At the time it annoyed me because it couldn't have been more different to what I asked for (hence the wry post above). But now I'm starting to think that maybe some kind of template approach might remove the need for a large list of, I'll admit, quite tedious and time consuming instructions.

Arggh! There I said it! I'm now a slave to the machine...

Tim
 
I tried it again tonite.

I took too long when playing and got a Network Timeout.

ChatGPT doesn't save sessions as far as I'm aware so I just ended it.

One more limitation of the Virtual GM - you have to have a consistent Internet Connection if you want to play a long game..
 
Now add Apple Vision and the soon to be developed Apple Haptic Playsuit (not really a thing) and the Chat AI will generate your whole life in any setting...

welcome to the Matrix cloudlings
 
TimInWharfedale said:
SS - it did feel exactly like that.

I omitted my response in post above . I did not wish to sully anybody's finer sensibilities. Suffice to say, my observations were not as measured as they were in the film.

After that it was surprisingly happy to carry on with whatever "tedious and time consuming" guff I cared to inflict on it. However, I took it as a warning of possible things to come if I pushed it too far - lol.

On a more practical note...

The AI did, as I mentioned, offer me a template to provide what I wanted instead of executing all the rules I provided. At the time it annoyed me because it couldn't have been more different to what I asked for (hence the wry post above). But now I'm starting to think that maybe some kind of template approach might remove the need for a large list of, I'll admit, quite tedious and time consuming instructions.

Arggh! There I said it! I'm now a slave to the machine...

Tim
Click to expand...
The little guys in the red coveralls will be around soon to install your new interface.
 
I have asked ChatGPT to provide scenarios. I wanted a scenario set on Mars, I wanted a medical mystery and I wanted it set in "Space City" - whatever that is. Please note that my layout for a scenario is:
  1. Individualistic, in other words non-standard. You might want to do it differently. That especially include adding more detail. I deliberately restricted this - but ChatGPT loves to give you detail, so fill yer boots!.
  2. The emphasis is on solo play so, again, you might want it to be a little more free form, and with fewer dice rolling, than I have set it out.
Anyway this is what it did as a single example. It came up with everything, no prompts from me (other than take place on Mars, medical mystery, space city etc.) though I did specify the layout and formatting (which was much harder to control than the ideas):

RED SANDS INTRIGUE
A mysterious illness plagues the citizens of the space city on Mars, and it's up to the team of adventurers to uncover the cause and find a cure before it spreads further.

POSSIBLE CAUSES – d6:
(1) Unlikely: The illness is caused by an ancient Martian curse that was triggered by recent excavation activities.
(2–3) Possible: The illness is a result of a biological experiment gone wrong in one of the medical research facilities.
(4–6) Likely: The illness is a new strain of a known virus that has mutated and become highly contagious.

PLAUSIBLE ACTIONS:
A: Investigate the recent excavation sites and search for any artifacts or symbols associated with curses.
B: Gather information from the medical research facilities and identify any ongoing experiments or accidents.
C: Analyze the virus samples and study its genetic makeup to understand its origins and potential weaknesses.
D: Interview the affected individuals and track their movements to identify any common patterns or sources of exposure.
E: Secure the infected area and implement quarantine measures to prevent further spread of the illness.

RANDOM COMPLICATIONS – d6:
1 – Individual: One of the adventurers starts exhibiting symptoms of the illness, adding urgency to finding a cure.
2 – Group: A rival team of researchers interferes with the investigation, seeking to claim credit for the cure.
3 – Place: The main medical research facility is compromised, making it difficult to access crucial data and equipment.
4 – Thing: A stolen vial of the virus is found in the possession of a criminal organization, who intend to weaponize it.
5 – Incident: A riot breaks out among the affected citizens due to fear and desperation, causing chaos and hindering the investigation.
6 – Idea: The illness turns out to be a deliberate act of sabotage by a disgruntled former employee seeking revenge.

It isn't perfect, but it's not bad either in my view. Crucially it would take me ages to come up with something like this.

What do you think?

Tim
 
TimInWharfedale said:
I have asked ChatGPT to provide scenarios. I wanted a scenario set on Mars, I wanted a medical mystery and I wanted it set in "Space City" - whatever that is. Please note that my layout for a scenario is:
  1. Individualistic, in other words non-standard. You might want to do it differently. That especially include adding more detail. I deliberately restricted this - but ChatGPT loves to give you detail, so fill yer boots!.
  2. The emphasis is on solo play so, again, you might want it to be a little more free form, and with fewer dice rolling, than I have set it out.
Anyway this is what it did as a single example. It came up with everything, no prompts from me (other than take place on Mars, medical mystery, space city etc.) though I did specify the layout and formatting (which was much harder to control than the ideas):

RED SANDS INTRIGUE
A mysterious illness plagues the citizens of the space city on Mars, and it's up to the team of adventurers to uncover the cause and find a cure before it spreads further.

POSSIBLE CAUSES – d6:
(1) Unlikely: The illness is caused by an ancient Martian curse that was triggered by recent excavation activities.
(2–3) Possible: The illness is a result of a biological experiment gone wrong in one of the medical research facilities.
(4–6) Likely: The illness is a new strain of a known virus that has mutated and become highly contagious.

PLAUSIBLE ACTIONS:
A: Investigate the recent excavation sites and search for any artifacts or symbols associated with curses.
B: Gather information from the medical research facilities and identify any ongoing experiments or accidents.
C: Analyze the virus samples and study its genetic makeup to understand its origins and potential weaknesses.
D: Interview the affected individuals and track their movements to identify any common patterns or sources of exposure.
E: Secure the infected area and implement quarantine measures to prevent further spread of the illness.

RANDOM COMPLICATIONS – d6:
1 – Individual: One of the adventurers starts exhibiting symptoms of the illness, adding urgency to finding a cure.
2 – Group: A rival team of researchers interferes with the investigation, seeking to claim credit for the cure.
3 – Place: The main medical research facility is compromised, making it difficult to access crucial data and equipment.
4 – Thing: A stolen vial of the virus is found in the possession of a criminal organization, who intend to weaponize it.
5 – Incident: A riot breaks out among the affected citizens due to fear and desperation, causing chaos and hindering the investigation.
6 – Idea: The illness turns out to be a deliberate act of sabotage by a disgruntled former employee seeking revenge.

It isn't perfect, but it's not bad either in my view. Crucially it would take me ages to come up with something like this.

What do you think?

Tim
Click to expand...
Wow! That's kind of amazing!

If Curses are ok in the Campaign, well and good. But if not, I was thinking of a few scientific possibilities that the Curse actually is.

1. Prehistoric (fill in the blank).
2. Alien (fill in the blank).
3. Weaponized (fill in the blank).

Whichever it is, or even a combination, the Ancient Martians were unable to eradicate it, but were able to contain it. The containment has recently been breached by (fill in the blank) and now we need to figure out how it escaped, what the Martians knew, and what can we do about it now.
 
Thank you to all the testers who helped out at tonight's session.

Here's a recap of what happened tonight:

WOW.. just WOW

Tonight's tests were pretty good and helped us identify some weak spots that need to be addressed in v0.2.

We were amazed that the AI could produce a basic visual layout of the ship using ASCII.

We managed to make our way to the bridge, where we found it pretty much destroyed - the windows are all frosted over and the main viewscreen was inoperable and heavily damaged.

After making some repairs to the power systems feeding the bridge, we managed to regain helm control and restore the security station. We were able to activate the skin cameras and inspect the exterior for damage. The shuttle disposition was unknown but no lifeboats had been launched.

Upon examination of the engines, we realized more repairs were needed. We made our way back to Engineering and repaired the engines.

Going back to the bridge, we brought the engines online and set a course away from the anomaly.

Before leaving, we decided to try to find out what happened to the crew, so we launched a science probe and gathered a bunch of info about the anomaly.

Feeling like the adventure was at a turning point, we decided to stop here for now.
 
Spinward Scout said:
I tried it again tonite.

I took too long when playing and got a Network Timeout.

ChatGPT doesn't save sessions as far as I'm aware so I just ended it.

One more limitation of the Virtual GM - you have to have a consistent Internet Connection if you want to play a long game..
Click to expand...
it does save sessions just come back to the game at any time - if you get an error message just reload the response or refresh the page
came back to the demo game the second night - continued from where we left off. played about 4 hrs (total of 8)
did experience the problem mentioned but regenerate response or refresh took care of it for us.
 
CyborgPrime said:
it does save sessions just come back to the game at any time - if you get an error message just reload the response or refresh the page
came back to the demo game the second night - continued from where we left off. played about 4 hrs (total of 8)
did experience the problem mentioned but regenerate response or refresh took care of it for us.
Click to expand...
Thanks! I'll remember that for next time.

TimInWharfedale said:
SS - it did feel exactly like that.
Click to expand...
The more I re-read this, the more it scares me.
 
Last edited:
BW - I think your approach to the concept of a "curse" is the way to go. It could literally be a curse of course, but that's more pirates in the Caribbean I think. For me, like yourself, it's more representative of something bad that won't go away. The point is I wouldn't have thought of that myself. At worst, I can ignore the output and manually replace it with something more plausible. While at best it would get me thinking what kind of "curse" it might be. There's no downside to this (unless the AI starts to take over - lol) because you can ignore the output. This I did many times. I just asked it to regenerate and off it goes with a new one - job of a couple of seconds. If it develops a preoccupation with "curses" (it does have a tendency to fixate like this). You might have to add an instruction that says something like:

"None of the possible causes can be a curse". This can be annoying to have to do this. Alternatively you can actually tell it to "Stop using a curse as a possible cause" as a text response.

Unfortunately ChatGPT is very absent minded (or willfully uncooperative? - lol) and neither a rule or a text reminder is likely to work for long. If I tell it to something that it doesn't want to, it tends to do it for awhile and then, when it thinks I've forgotten, do what it did before. So if it wants to talk about curses then it probably will. You have to hope that it doesn't do it too long. I think of ChatGPT not as a tool (eg: like a mobile phone) but more as a helpful (wayward?) side kick (eg: like Short Round). By that I mean you can tell it to do something and give it very clear instructions. Sometimes it does what you want, sometimes it doesn't and gives you something it likes better instead. There really are two of you in this, even if the AI is a junior partner (for the moment) in this activity.

If anybody is interested I've attached the instructions and template I used to create the scenario I posted earlier. Notes:
  1. The instructions seem quite long to me for what is required (but who knows?). My instincts tell me that if I could write this more cleverly I could do it shorter. For instance, before using Markdown Syntax the instructions were both longer and less likely to be executed. So it can always be improved.
  2. In my original post, I referred to the format being Traveller non-standard. It is fairly easy to create a more conventional layout (eg: Supplement 16 - Adventure Seeds or similar). Just alter the instructions to suit.
  3. Text in red in boxes is a reminder to me to change these variables at run time if needed.
  4. You might notice that some instructions are repeated. This adds to the length of the whole set of instructions, but it is an attempt to get ChatGPT to follow my instructions where there seem to be temptations to do something else.
  5. As mentioned, I've found it helpful to use Markdown Syntax (see here) which it responds to better than written text when it comes to formatting.
  6. I've broken the instructions into sets, each relating to a different aspect of the response. It helps me keep things in order and ChatGPT carry them out as well. Although it can piece things together better than most humans do if you want to do things in a jumbled and disorganized way.
  7. I think it will execute the instructions in the order written. So if everything is written so that there are no contradictions there is no problem. But if there are contradictions then I think it will follow the first instruction over any later ones.
  8. If your instructions are unclear, or just misspelled, it will ignore them. Incidentally, it is happy with both American English and real English (lol). So you don't need to Anglicize or Americanize your spelling.
  9. There is a mention of "background" (at the very end of the instructions). You probably don't need this if you are using the OTU, but mine is a 22nd century version restricted to the solar system (eg: no FTL, Psionics, anti-gravity etc.). so I had to describe all this in a separate set of background notes. If YOU have anything you want to exclude (or extra that you want to include) then background notes are really helpful for ChatGPT, so it knows what it should work with and what to ignore.
I hope this is interesting. See attachment for the actual Instructions and template I gave it create the sample scenario. You can try it out for yourself by COPYing and PASTEing it into ChatGPT. Why not change a few variables to see how it works for you? If you can improve on it's reliability, inventiveness or anything else (especially simplifying the instructions!) then please let me know.

Have fun!

Tim
 

Attachments

  • SpecimenAIScenarioTemplate.docx
    15.2 KB · Views: 7
All - here's another specimen, same variables as before (ie: Mars, Space City, 22nd Century, not OTU etc.).

ChatGPT invented "Red Pharma", which I quite like. It goes well with the Traveller concept of corporate power.

Several parts of the scenario need further fleshing out, especially the complications. But that's exactly how I want it. I can add a plug in elements (eg: a faction, a secret society, a rival team, a rival corporation etc.) from what is both relevant and known to the adventurers:

RED PHARMA CONSPIRACY​

A Mysterious Epidemic Strikes the Mars City​

CAUSES – d6:
(1) Unlikely: The epidemic is caused by a rare alien virus.
(2–3) Possible: The epidemic is the result of a faulty batch of a popular Martian medicine.
(4–6) Likely: The epidemic is intentionally caused by a rival pharmaceutical corporation.

ACTIONS:
A: Investigate the source of the epidemic by examining medical records.
B: Interview the affected patients and their families for commonalities.
C: Gather evidence from the manufacturing facility where the medicine was produced.
D: Infiltrate the rival corporation to uncover their motives and involvement.
E: Develop an antidote or treatment to combat the effects of the epidemic.

COMPLICATIONS – d6:
1 – Individual: One of the adventurers becomes infected with the alien virus.
2 – Group: A faction within the Mars City government tries to cover up the conspiracy.
3 – Place: The manufacturing facility is heavily guarded, making investigation challenging.
4 – Thing: The antidote requires a rare Martian plant that is on the brink of extinction.
5 – Incident: A rival team of adventurers is also investigating the conspiracy, creating competition and potential conflicts.
6 – Idea: The rival corporation is actually being manipulated by a secret society aiming to destabilize Mars' medical infrastructure.

Creating these can be addictive (but that's what ChatGPT wants so that you become dependent - lol).

Tim
 
You must log in or register to reply here.

Similar threads

Commander Truestar
General I would like your thoughts on this adventure idea...
2 3 4
Replies
69
Views
1K
Grav_Moped
Grav_Moped
Commander Truestar
  • Article
General Question, I have a base image of an Imperial T Class Cruiser....
Replies
5
Views
165
Grav_Moped
Grav_Moped
W
EURISKO LIVES!
Replies
1
Views
618
MichaelSTee
M
Tobias
  • Poll
General Battledress: Suits or Mecha?
5 6 7
Replies
134
Views
2K
aramis
aramis
Marcatlas
Solid Air
Replies
7
Views
212
Tobias
Tobias
Back
Top