核心概念
LLMR enables real-time creation and modification of interactive 3D scenes using Large Language Models.
要約
LLMR framework overview:
Authors from MIT, Microsoft, and Rensselaer Polytechnic Institute collaborated on the research.
LLMR leverages novel strategies for creating interactive Mixed Reality experiences.
Key Modules:
Planner breaks down user requests into manageable subtasks.
Scene Analyzer provides a summary of the virtual scene for downstream processing.
Builder generates code based on user input and scene summary.
Inspector checks generated code for errors before execution.
Skill Library retrieves relevant skills needed for implementation.
Use Cases:
Game design, creativity, accessibility features, adaptive interfaces, remote assistance, and planning simulations demonstrated.
Installation & Compatibility:
LLMR can be easily added to existing Unity scenes with OpenAI API key requirement.
Memory Management:
Three memory modes implemented for efficient token usage and performance optimization.
統計
LLMRは、標準のGPT-4に比べて平均エラー率を4倍改善しました。