Research map / Large language models / Agents, code and math
MarioGPT: Open-Ended Text2Level Generation through Large Language Models
Shyam Sudhakaran, Miguel González-Duque, Matthias Freiberger, Claire Glanois, Elias Najarro, Sebastian Risi
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| fine-tune pretrained GPT2 weights rather than train the level model from scratch (ablation) | MarioGPT · Mario level validation set | Tile accuracy: 31% -> 93% | Improved |
| fine-tune pretrained GPT2 weights rather than train the level model from scratch (ablation) | MarioGPT · Mario level validation set | Path accuracy: 23% -> 91% | Improved |
| replace pretrained-model fine-tuning with an adapter layer (ablation) | MarioGPT · Mario level validation set | Tile accuracy: 93% -> 21% | Hurt |
| replace pretrained-model fine-tuning with an adapter layer (ablation) | MarioGPT · Mario level validation set | Path accuracy: 91% -> 11% | Hurt |
| condition level generation on natural-language prompts | MarioGPT · Prompt-conditioned Mario levels | Prompt accuracy: blocks: 92% | Improved |
| condition level generation on natural-language prompts | MarioGPT · Prompt-conditioned Mario levels | Prompt accuracy: enemies: 68% | Mixed |
| increase sampling temperature (ablation) | MarioGPT · Generated Mario levels | Sample diversity | Mixed |
| increase context length beyond 700 (ablation) | MarioGPT · Mario level generation | Performance | No clear effect |
| use novelty search instead of random-prompt sampling to select diverse levels | NS-MarioGPT · Generated Mario levels | Level diversity | Improved |
| add novelty search with predicted-path behavior characteristics | NS-MarioGPT · Novelty-search archive | Possible coordinates covered: 78% | Improved |
| use model-generated paths as path guidance for level generation | MarioGPT · Playable Mario levels | Path MAE: 1.15 | Improved |
Closest papers by meaning
- Moonshine: Distilling Game Content Generators into Steerable Generative Models (AAAI 2025 · similarity 0.84)
- GameGen-X: Interactive Open-world Game Video Generation (ICLR 2025 · similarity 0.83)
- Procedural Level Generation with Diffusion Models from a Single Example (AAAI 2024 · similarity 0.80)
- GameTraversalBenchmark: Evaluating Planning Abilities Of Large Language Models Through Traversing 2D Game Maps (NeurIPS 2024 · similarity 0.79)
- Fantastic Copyrighted Beasts and How (Not) to Generate Them (ICLR 2025 · similarity 0.79)
- SceneX: Procedural Controllable Large-Scale Scene Generation (AAAI 2025 · similarity 0.78)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
