Concise prompting makes LLM models cheaper

A study done by Johns Hopkins University on Concise Chain-of-Thought (CCoT) Prompting revealed that concise prompts lead to:

•⁠ ⁠Shorter Responses: CCoT nearly halves the response length for GPT-3.5 and GPT-4.

•⁠ ⁠Cost Efficiency: This method can reduce per-token costs by up to 23.49%, leading to significant savings.

•⁠ ⁠Consistent Performance: There's no major drop in problem-solving accuracy, except for some reduction in math problem accuracy for GPT-3.5.

These research findings have direct implications for AI systems engineers. By implementing concise prompting, they can significantly optimize the efficiency and cost-effectiveness of their AI solutions without compromising on problem-solving accuracy.

If you're interested in the details, check out the complete research -> https://arxiv.org/pdf/2401.05618