Green prompt engineering for sustainable generative AI

Area of Science:

• Artificial Intelligence
• Computer Science

Background:

• Prompt engineering is crucial for controlling large language model (LLM) outputs.
• High computational costs and energy consumption are significant challenges in generative AI.
• Suboptimal prompts increase LLM inference iterations, escalating energy use and carbon footprint.

Purpose of the Study:

• To propose practices and guidelines for energy-efficient prompt design.
• To minimize reiterations required for desired LLM responses.
• To enhance the sustainability of generative AI applications.

Main Methods:

• Developed a series of best practices for prompt engineering.
• Conducted empirical evaluations across various LLMs and scenarios.
• Measured energy consumption and greenhouse gas emissions.

Main Results:

• Applied best practices reduced energy consumption by 32-48%.
• Significant decrease in operational greenhouse gas emissions observed.
• Validated effectiveness across diverse LLMs and test cases.

Conclusions:

• Proposed best practices enhance LLM prompt inferencing energy efficiency.
• These guidelines can be integrated into generative AI design frameworks.
• Advocates for sustainable and effective deployment of generative AI.