GPTs powered Apocalypse game need feedback

Hello there, merry christmas

Im working on an personal experiment on GPT powered games:
a MUD in an apocalypse setting, needs your feedbacks!!

Key research points:

  1. Content generation consistency
    How stable is the content generation?
  2. Complex rule set understanding - max out instruction set
    Does GPTs follow the game structure and rules correctly?
  3. Stable code execution for determined game mechanisms.
    Such as action difficulty evaluation, dice roll, outcome calculation etc.
  4. External API connection (send score to external leader-board)
  5. NPC interaction with NPCs remember and understand the world, looking into connecting with Inworld AI. api,

Intro Video

Play & provide feedback at

You can also add me on LinkedIn


Do I have to provide email address to create game profile?

Lovely. Im building a similar thing. Hiw do you controle the lore of the game. Description, knowledge or custom API? Im struggeling with structured imtroduction of user-generated content. Is it open-source. Would live to team up.

I need a unique id to group multiple score submission into one

1 Like

In the instruction you can ask GPT to follow a constructed steps.
The lore you can put into knowledge,
then you can put ask GPT to search lore in knowledge base.

The concept looks great. I really like the idea of using GPTs to program a proper game with a structured game loop and a fairly complex ruleset, going beyond some freeform RPG or narrative. There aren’t many out there.

I have been also creating a structured game with GPTs, but not to this degree of complexity. I am truly impressed that you managed to get it to work & follow the rules consistently!

Unfortunately, I didn’t manage to get your GPT to work past the first Day turn as I keep getting “network errors” or the infamous “There was an error generating a response”. This might not be because of your GPT, but just ChatGPT being overloaded these days… I will try again at another time.

Anyhow, what I have seen so far seems truly impressive in terms of the amount of “programming” and game logic that you managed the GPT to follow. I frankly didn’t know it was possible to this extent; I imagined ChatGPT would keep losing track of things.

How much did you have to tweak your instructions to get GPT to follow your rules consistently? And did you find some specific way of prompting that you find particularly effective?

I used up the entire instruction set.

Some technique I use for example
I ask GPT to follow constructed steps 1, 2, 3, 4 etc
I use symbols to group the instructions together.
[1.[ 1.1 1.2 1.3 ]]
And for some calculation I hard code python script and upload into the knowledge base, then tell GPT to feed its input.

A part from that just a lot of tweaking. Yeah the “There was an error generating a response” happens all the time, i doubt it has anything to do with instructions.