Level Generation Through Large Language Models

Graham Todd; Sam Earle; Muhammad Umair Nasir; Michael Cerny Green; Julian Togelius

doi:10.1145/3582437.3587211

Level Generation Through Large Language Models

Graham Todd, Sam Earle, Muhammad Umair Nasir, Michael Cerny Green, Julian Togelius

Research output: Chapter in Book/Report/Conference proceeding › Conference contribution

Abstract

Large Language Models (LLMs) are powerful tools, capable of leveraging their training on natural language to write stories, generate code, and answer questions. But can they generate functional video game levels? Game levels, with their complex functional constraints and spatial relationships in more than one dimension, are very different from the kinds of data an LLM typically sees during training. Datasets of game levels are also hard to come by, potentially taxing the abilities of these data-hungry models. We investigate the use of LLMs to generate levels for the game Sokoban, finding that LLMs are indeed capable of doing so, and that their performance scales dramatically with dataset size. We also perform preliminary experiments on controlling LLM level generators and discuss promising areas for future work.

Original language	English (US)
Title of host publication	Proceedings of the 18th International Conference on the Foundations of Digital Games, FDG 2023
Editors	Phil Lopes, Filipe Luz, Antonios Liapis, Henrik Engstrom
Publisher	Association for Computing Machinery
ISBN (Electronic)	9781450398565
DOIs	https://doi.org/10.1145/3582437.3587211
State	Published - Apr 12 2023
Event	18th International Conference on the Foundations of Digital Games, FDG 2023 - Lisbon, Portugal Duration: Apr 11 2023 → Apr 14 2023

Publication series

Name	ACM International Conference Proceeding Series

Conference

Conference	18th International Conference on the Foundations of Digital Games, FDG 2023
Country/Territory	Portugal
City	Lisbon
Period	4/11/23 → 4/14/23

Keywords

language models
procedural content generation
sokoban
transformers

ASJC Scopus subject areas

Human-Computer Interaction
Computer Networks and Communications
Computer Vision and Pattern Recognition
Software

Access to Document

10.1145/3582437.3587211

Cite this

Todd, G., Earle, S., Nasir, M. U., Green, M. C., & Togelius, J. (2023). Level Generation Through Large Language Models. In P. Lopes, F. Luz, A. Liapis, & H. Engstrom (Eds.), Proceedings of the 18th International Conference on the Foundations of Digital Games, FDG 2023 Article 70 (ACM International Conference Proceeding Series). Association for Computing Machinery. https://doi.org/10.1145/3582437.3587211

Level Generation Through Large Language Models. / Todd, Graham; Earle, Sam; Nasir, Muhammad Umair et al.
Proceedings of the 18th International Conference on the Foundations of Digital Games, FDG 2023. ed. / Phil Lopes; Filipe Luz; Antonios Liapis; Henrik Engstrom. Association for Computing Machinery, 2023. 70 (ACM International Conference Proceeding Series).

Research output: Chapter in Book/Report/Conference proceeding › Conference contribution

Todd, G, Earle, S, Nasir, MU, Green, MC & Togelius, J 2023, Level Generation Through Large Language Models. in P Lopes, F Luz, A Liapis & H Engstrom (eds), Proceedings of the 18th International Conference on the Foundations of Digital Games, FDG 2023., 70, ACM International Conference Proceeding Series, Association for Computing Machinery, 18th International Conference on the Foundations of Digital Games, FDG 2023, Lisbon, Portugal, 4/11/23. https://doi.org/10.1145/3582437.3587211

@inproceedings{dee2b2c6725f4b1ca60301f9936ae495,

title = "Level Generation Through Large Language Models",

abstract = "Large Language Models (LLMs) are powerful tools, capable of leveraging their training on natural language to write stories, generate code, and answer questions. But can they generate functional video game levels? Game levels, with their complex functional constraints and spatial relationships in more than one dimension, are very different from the kinds of data an LLM typically sees during training. Datasets of game levels are also hard to come by, potentially taxing the abilities of these data-hungry models. We investigate the use of LLMs to generate levels for the game Sokoban, finding that LLMs are indeed capable of doing so, and that their performance scales dramatically with dataset size. We also perform preliminary experiments on controlling LLM level generators and discuss promising areas for future work.",

keywords = "language models, procedural content generation, sokoban, transformers",

author = "Graham Todd and Sam Earle and Nasir, {Muhammad Umair} and Green, {Michael Cerny} and Julian Togelius",

note = "Publisher Copyright: {\textcopyright} 2023 ACM.; 18th International Conference on the Foundations of Digital Games, FDG 2023 ; Conference date: 11-04-2023 Through 14-04-2023",

year = "2023",

month = apr,

day = "12",

doi = "10.1145/3582437.3587211",

language = "English (US)",

series = "ACM International Conference Proceeding Series",

publisher = "Association for Computing Machinery",

editor = "Phil Lopes and Filipe Luz and Antonios Liapis and Henrik Engstrom",

booktitle = "Proceedings of the 18th International Conference on the Foundations of Digital Games, FDG 2023",

}

TY - GEN

T1 - Level Generation Through Large Language Models

AU - Todd, Graham

AU - Earle, Sam

AU - Nasir, Muhammad Umair

AU - Green, Michael Cerny

AU - Togelius, Julian

PY - 2023/4/12

Y1 - 2023/4/12

N2 - Large Language Models (LLMs) are powerful tools, capable of leveraging their training on natural language to write stories, generate code, and answer questions. But can they generate functional video game levels? Game levels, with their complex functional constraints and spatial relationships in more than one dimension, are very different from the kinds of data an LLM typically sees during training. Datasets of game levels are also hard to come by, potentially taxing the abilities of these data-hungry models. We investigate the use of LLMs to generate levels for the game Sokoban, finding that LLMs are indeed capable of doing so, and that their performance scales dramatically with dataset size. We also perform preliminary experiments on controlling LLM level generators and discuss promising areas for future work.

AB - Large Language Models (LLMs) are powerful tools, capable of leveraging their training on natural language to write stories, generate code, and answer questions. But can they generate functional video game levels? Game levels, with their complex functional constraints and spatial relationships in more than one dimension, are very different from the kinds of data an LLM typically sees during training. Datasets of game levels are also hard to come by, potentially taxing the abilities of these data-hungry models. We investigate the use of LLMs to generate levels for the game Sokoban, finding that LLMs are indeed capable of doing so, and that their performance scales dramatically with dataset size. We also perform preliminary experiments on controlling LLM level generators and discuss promising areas for future work.

KW - language models

KW - procedural content generation

KW - sokoban

KW - transformers

UR - http://www.scopus.com/inward/record.url?scp=85153576891&partnerID=8YFLogxK

UR - http://www.scopus.com/inward/citedby.url?scp=85153576891&partnerID=8YFLogxK

U2 - 10.1145/3582437.3587211

DO - 10.1145/3582437.3587211

M3 - Conference contribution

AN - SCOPUS:85153576891

T3 - ACM International Conference Proceeding Series

BT - Proceedings of the 18th International Conference on the Foundations of Digital Games, FDG 2023

A2 - Lopes, Phil

A2 - Luz, Filipe

A2 - Liapis, Antonios

A2 - Engstrom, Henrik

PB - Association for Computing Machinery

T2 - 18th International Conference on the Foundations of Digital Games, FDG 2023

Y2 - 11 April 2023 through 14 April 2023

ER -

Level Generation Through Large Language Models

Abstract

Publication series

Conference

Keywords

ASJC Scopus subject areas

Access to Document

Other files and links

Fingerprint

Cite this