Skip to content

Latest commit

 

History

History
83 lines (58 loc) · 4.32 KB

README.md

File metadata and controls

83 lines (58 loc) · 4.32 KB
Raw

Creating MapleStory (odin based source) NPC Script AI using GPT4

This is a small technical demonstration showcasing the capabilities of GPT in relation to legacy MapleStory (odin-based source) scripts.

  • State based NPC script to stateless NPC script - These scripts are state-based and can often be laborious to write, and even more tedious to convert to a stateless alternative today (thousands upon thousands of them).
  • NPC story-writer + script-writer.

See it live here: https://chat.openai.com/g/g-P3LABcaro-maplestory-npc-scripter

  • 08/03/2023 -- With the release of gpt-3.5-turbo model today that makes it even better, I shall present you with the following that I have for a while.
  • 29/03/2023 -- updated for GPT4
  • 18/02/2024 -- An example has been included live. Link

Gif output

More examples:

Converting from state-based NPC script to stateless:

image

Creating a series of NPC scripts revolving around a single theme:

image

Creating complex Tic-Tac-Toe NPC script:

image

Have fun~! AI-ing~!!

Instructions to run

Installation & setup

  1. Install NodeJS 18: https://nodejs.org/en/
  2. npm install
  3. Create an account, and get an OpenAI API key
  4. Go to config.ts and enter the API key under OPENAI_API_KEY
  5. Place your NPC scripts under _input\npc\ folder

Start

  1. npx ts-node script.ts
  2. You should see the refractored files under _output\npc\

FYI

Limitations

  • You have to adjust the prompts according to your own MapleStory sources, it is not guaranteed to work as you may have different function names for the NPC conversation.
  • It may not work 100% of the time. Try to change the "temperature" parameters to 0.2 from 0 for a little added randomless to the output instead of a fixed-deterministic output.
  • ALWAYS test it, before going to production. Every single script.
  • Further optimisation is possible down the road when fine-tuning of a model is supported for GPT3.5/GPT4 and above. To save both cost and for higher accuracy.
  • It might not be able to refractor if the script's input size is > 32,768 tokens due to GPT4's current limitation.
  • It may sometimes hallucinate to create functions that lead to nowhere, I'm relatively sure GPT is trained on other MapleStory Odin-based sources out there on GitHub. It might be possible to solve this by showing it more examples of scripts refractored manually by a human vs its unrefractored counterpart when the limitations of 4096 tokens are lifted. Example:
function onUserEnter() {
    cm.say("I'm the head of this hair salon. If you have a #b#t5150053##k or a #b#t5151036##k allow me to take care of your hairdo. Please choose the one you want.");
}

function onUserExit() {
    cm.say("Thank you for visiting our salon. Come back soon!");
}

function onDisconnected() {
    cm.say("Oh no! You have disconnected. Please log back in and visit us again!");
}

How its done:

  1. [System] Prompts GPT what persona it needs to be, or to act as.
  2. [System] Tells GPT the stateless NPC functions available
  3. [User] Tells GPT to refractor the script to a stateless NPC script.
  4. [Assistant] Actually shows it a valid refractored example to guide its knowledge further.
  5. [User] Give it a real unconvered script, and ask for conversion.

Useful link