The growth of Large Language Models (LLMs) in daily operations necessitates rigorous evaluation methods. By introducing a multi-agent AI model, this paper presents preliminary results for assessing different LLMs. These agents execute code retrieval tasks across a range of language models and validates through a verification agent, providing insights into their comparative performances.

Key Points:

• A multi-agent AI model evaluates LLMs, comparing their performance systematically.
• Utilizes the HumanEval benchmark for the verification of generated code’s performance.
• Initial results show GPT-3.5 Turbo outperforms other models.

Further Insights:

This evaluation method has the potential to standardize LLM performance assessments, offering valuable metrics to guide further improvements in the development of these complex systems. Evaluating LLMs