Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Game Development as Human-LLM Interaction

Jiale Hong, Hongqiu Wu, Hai Zhao

TL;DR

ChatGE presents a framework that enables game development through Human-LLM interaction, addressing the steep learning curve of traditional engines. It combines a data synthesis pipeline with a three-stage progressive training strategy to align LLMs with joint script generation and coding tasks. A poker case study shows that ChatGE achieves high interaction quality and near-perfect code correctness, outperforming baselines and ablations. The work highlights practical potential for LLM-driven game development while acknowledging scalability and modality limitations as avenues for future research.

Abstract

Game development is a highly specialized task that relies on a complex game engine powered by complex programming languages, preventing many gaming enthusiasts from handling it. This paper introduces the Chat Game Engine (ChatGE) powered by LLM, which allows everyone to develop a custom game using natural language through Human-LLM interaction. To enable an LLM to function as a ChatGE, we instruct it to perform the following processes in each turn: (1) $P_{script}$: configure the game script segment based on the user's input; (2) $P_{code}$: generate the corresponding code snippet based on the game script segment; (3) $P_{utter}$: interact with the user, including guidance and feedback. We propose a data synthesis pipeline based on LLM to generate game script-code pairs and interactions from a few manually crafted seed data. We propose a three-stage progressive training strategy to transfer the dialogue-based LLM to our ChatGE smoothly. We construct a ChatGE for poker games as a case study and comprehensively evaluate it from two perspectives: interaction quality and code correctness.

Game Development as Human-LLM Interaction

TL;DR

ChatGE presents a framework that enables game development through Human-LLM interaction, addressing the steep learning curve of traditional engines. It combines a data synthesis pipeline with a three-stage progressive training strategy to align LLMs with joint script generation and coding tasks. A poker case study shows that ChatGE achieves high interaction quality and near-perfect code correctness, outperforming baselines and ablations. The work highlights practical potential for LLM-driven game development while acknowledging scalability and modality limitations as avenues for future research.

Abstract

Game development is a highly specialized task that relies on a complex game engine powered by complex programming languages, preventing many gaming enthusiasts from handling it. This paper introduces the Chat Game Engine (ChatGE) powered by LLM, which allows everyone to develop a custom game using natural language through Human-LLM interaction. To enable an LLM to function as a ChatGE, we instruct it to perform the following processes in each turn: (1) : configure the game script segment based on the user's input; (2) : generate the corresponding code snippet based on the game script segment; (3) : interact with the user, including guidance and feedback. We propose a data synthesis pipeline based on LLM to generate game script-code pairs and interactions from a few manually crafted seed data. We propose a three-stage progressive training strategy to transfer the dialogue-based LLM to our ChatGE smoothly. We construct a ChatGE for poker games as a case study and comprehensively evaluate it from two perspectives: interaction quality and code correctness.
Paper Structure (36 sections, 4 equations, 3 figures, 14 tables)

This paper contains 36 sections, 4 equations, 3 figures, 14 tables.

Table of Contents

  1. Introduction
  2. Related works
  3. AI for Games
  4. LLMs as Training Data Generators
  5. Progressive Training Strategy
  6. ChatGE
  7. Overview
  8. Formulation
  9. Data Generation
  10. Init pool
  11. Generate new pairs
  12. Generate interaction data
  13. Training Strategy
  14. Stage-1: Base Training
  15. Stage-2: Core Training
  16. ...and 21 more sections

Figures (3)

  • Figure 1: An illustration of the ChatGE process for a poker game. See Appendix \ref{['app-sec:example']} for a complete example.
  • Figure 2: ChatGE framework. The LLM processes the user's input in the orange stream, while simultaneously generating script in the yellow stream, code in the green stream, and interaction in the blue stream.
  • Figure 3: Data synthesis pipeline for game script-code pair and interaction generation.