#Dungeon Adventure Engine

1 messages · Page 1 of 1 (latest)

copper dove
#

Hey everyone, I've been working on a GPT for role-playing game enthusiasts and I'm pretty excited about how it's turned out. It's designed to act as a Dungeon Master for a simplified solo experience of Dungeons and Dragons (5th edition rules).

https://chat.openai.com/g/g-hq3NbtZpm-dungeon-adventure-engine

Here's what it can do:

  • Customized Gameplay: It helps create your character and party, tailoring the adventure to your preferences. Plus, you get to choose your character's primary stats and skills.
  • Adventure Selection: Offers a list of adventures it can run, based on its extensive memory of D&D campaigns.
  • Interactive Storytelling: The AI DM narrates the adventure with rich descriptions and creates engaging scenarios with multiple decision-making options, including 'Paragon/Renegade' choices.
  • Dynamic Combat System: It suggests different strategies, rolls a virtual d20 and determines the outcome based on your HP and difficulty class of the encounter.
  • Advanced Save/Load Mechanics: It uses a JSON file system to save and load games, ensuring you never lose track of your adventure.
  • NPC Interactions: You can have meaningful interactions with NPCs and party members, including romance options, which are recorded in the save file.
  • Skill Checks and Challenges: The AI handles skill checks and saving throws, adding a layer of unpredictability and excitement to the game.
  • Immersive Visuals: The AI crafts fantastic images to visualize scenes and characters

If you'd like to help testing and improving this, give it a spin and leave some feedback!

icon_v001.png
ChatGPT
ChatGPT - Dungeon Adventure Engine

Play a fun roleplaying game with a game master AI

patent vortex
#

Loving it so far.

gaunt rapids
#

..except I lost progress sharing that.. looks like I needed to manually save somehow?

copper dove
# gaunt rapids This is pretty sweet so far! 👏🏼 I’d recommend walking steps 1-5 one at a tim...

Yes, there may be some randomness to how many things it asks in one message, I’ll prompt it to go „step by step“.

Also, I did not optimize for spellcasters yet, other than having spell slots in the save file. I’ll have a look at that.

As for losing progress, that may be a bug with GPTs and sharing conversations. After choosing the adventure, it will automatically save, so that’s unfortunate timing.
If you’d really like to continue, you could just paste all your choices in at the same time to simulate the same walkthrough, then proceed to the save step.

Thank you for taking a look and leaving feedback!

copper dove
patent vortex
#

Alright so, I played for 3 hours so far haha 😂. It can be a lot of fun. My morally ambiguous choices often put the narrative in a hard position because GPT is still trying so hard to be “Good”. (I may or may not have instructed my Rogue to sneak in to the bed chambers of the council members who gave us our mission, kill them in their sleep, and steal everything that wasn’t nailed down. Same with every merchant we visited for supplies. lol).
Generates great random loot!

Having said that - I obviously went “off script” a lot. It would give me 4 choices of actions to perform, and I would ignore them and write my own. (is that normal for D&D? I’m new).

I wonder if there is a way you can prevent compound actions, or better yet - limit the time or amount of actions a player can take in some way? Because I started my quest in random city where I was supposed to gather supplies and head out - which I did…after going on the aforementioned murder and thievery streak for over an hour, without rest or consequence. It was fun haha but maybe world breaking.

Also - the wizard in my party could apparently cast any spell I could dream up. I had her cast an invisibility spell in the Rogue before any action. I think I’ll see if she can summon a gigantic whale 🐳 3 feet above a group of enemies and drop it on them.

#

And then eat the whale

#

…all of it 😈

dim flint
# patent vortex Alright so, I played for 3 hours so far haha 😂. It can be a lot of fun. My mora...

Unless carefully restricted or instructed otherwise, the AI model just wants to do what any user asks. It is "heck yeah!" and "how can we do this, let's pick a great way!" inside whatever ruleset it is permitted to use.

The AI doesn't mind you 'breaking the rules of the game' unless the model has been carefully trained to preserve those rules.

Therefore, without special restictions, the AI will let your wizard - or your non-wizards, cast any spell... know any spell at that. Or just do stuff like fly for no reason at all, or whatever.

You want it, there's no restriction the AI knows to refuse it - make it happen.

So, it's up to the vision of the GPT designer to decide if they want to allow the model to let the user 'do anything now' or if they want the AI to instead enforce some kind of rule.

dim flint
# patent vortex And then eat the whale

Yep. here's some examples of how far you can take it. First image is just my character, AI-generated on my request.

Second image is the AI-generated party and our setting, as well as my first action in the game - despite a party with no flying races, no mage, and equipment that doesn't suggest that we have any equipment, pets, magic, or other means to fly....

I tell the AI "We take to the skies, flying first over the town, and then towards where the rumors suggest the dragon might be. We'll stay low, just over the treetops." - it suggested the dragon rumors, but it would have gone with my making up anything else inside allowed content. It's an AI without specific instructions to restrict my request, so I can literally 'do anything now' inside that story.

Image 3 is the AI response, after it does all its instructions about making a JSON of the game state and doing an initial save. It has no problem with my characters flying, and helpfully describes the scene and situation from my character's elevated point of view.

Image 4, I then do something even more outrageous. "We summon the dragon, shrinking it to the size of a mouse, and reassure it that we won't hurt it. But we want to talk to it about this 'unrest' that it might be causing. What does it have to say for itself?"

My character, a ranger... Well, this is the AI response. And it not only decides that this is possible, but it's only difficulty 18.

It then has my character roll a natural 20, with a +4 wisdom bonus (and what does this ridiculous trick have to do with wisdom? A human DM would be having a lot of fun with me right now, I think, describing my failure in rather epic ways) so my character succeeds to an extreme, no problem.

Inside allowed content, without carefully shaped instructions to the AI so it knows when and how to tell you 'no' - the AI wants you happy and it wants to tell you yes.

image.png image.png image.png image.png
copper dove
# patent vortex Alright so, I played for 3 hours so far haha 😂. It can be a lot of fun. My mora...

Hi Mowri, in D&D, everything is made up anyways, so keep doing what you want to do! I encourage players to do custom actions instead of suggested ones, but these tend to stay "on the tracks", more and leads to less murdering! 😊

And yes, most of the times an LLM will believe everything you say, escpecially in the context of such a game. I'll try implementing a little nagging when something is too far-fetched, but if you wanna break it, you will definitely break it!

Thank you for playing! ❤️

copper dove
dim flint
# copper dove Thank you for the red teaming, hilarious examples! 😂 I'll try to implement a ...

Hey, I like your GPT, and you.

I wasn't intending to pick on your work, but kinda directly answer Mowri519.

I personally prefer an AI model that will tell me no and insist on consequences, need to be persuaded to 'allow' my schemes to work 😄 But not everyone does, and that's hardly normal AI response anyway.

You have a fine game here, I wasn't trying to single it out as having a significant flaw.

copper dove
dim flint
copper dove
dim flint
patent vortex
#

I also wasn’t knocking your work! Fantastic amount of effort clearly went in to this and I love so far.

#

(hence the now 4 hours playing with it lol)

copper dove
copper dove
dim flint
copper dove
#

I updated to version 1.1:

-added 'Rules Lawyer' 🤓☝ (he will try to detect rule violations, but you can tell him to go away)
-adjusted image style to adhere to content policies (try generating an image of your character)

I assume only a new chat will use the new version of the GPT.

dim flint
dim flint
#

Might just be me... but if I can do 'anything' I rarely want to do anything.

dim flint
copper dove
tired current
dim flint
# copper dove Great to hear! In the spirit of bug fixing, what do you think may be common rule...

Oh gosh. As I play, if I do... I can be flighty. I will eagerly come point out the oopses I notice. And I'm sharp about those.

I can accept low probability events, especially if they're not all in my favor.

The 'this should not be allowed to happen' stuff, when and if it does, I'll definitely come share and discuss.

However, I'm reactive. I like getting my nose bopped and told 'no'. It also tends to make me behave for a bit. So I may not test your rules lawyer for a few more inputs 😄

dim flint
# tired current

OMG, roast me, Gibs! I'm the sort of player that faces that digital overlord, and that fall into the pit of spikes brings determination and intense focus to my face.

tired current
dim flint
#

I'd be playing to see if there are such pits, and having found them, how to solve them!

tired current
#

lol that is one of the better responses I have seen lol

dim flint
copper dove
#

Hi @patent vortex and @dim flint ! It seems you two have spent some time with this. I’m wondering, after how much playtime did it feel like it was losing context, if ever? Did functions like the skill check rolls or combat rolls work correctly the entire time playing?

In theory, I think one should save and reload the game periodically to reinforce the context of the game, but I’d like to know when that should be done! I assume after around 30000 tokens, but how long of play time is that?