This research introduces a novel MPC framework that transforms LLMs into powerful planning systems by treating them as implicit cost function minimizers.

• Demonstrates that LLMs with planning prompts naturally function as predictive controllers
• Unifies various prompting techniques under a single mathematical framework
• Shows measurable performance improvements over traditional few-shot prompting on planning benchmarks
• Provides a structured approach to optimize LLM planning abilities

For engineers, this research offers a systematic way to enhance LLM planning capabilities using established control theory principles, bridging the gap between traditional engineering approaches and modern AI systems.

LLMPC: Large Language Model Predictive Control