Thirteen LLMs asked the same 76 sourced questions, with and without web search (llmquorum.aidataforager.com)

In a recent evaluation, thirteen large language models (LLMs) were tested with a set of 76 sourced questions, comparing their performance with and without the ability to access web searches. The results revealed a significant improvement when models utilized web search, with paired comparisons showing that the models that could search answered more questions correctly compared to when they were restricted to their internal knowledge. This demonstrates the substantial impact that web access has on the accuracy of AI model outputs. The study specifically highlighted metrics such as accuracy rates, categorizing responses into groups like "Correct," "Outdated," and "Hallucinated." Across the board, models leveraging web search achieved higher accuracy rates, underscoring the importance of real-time information retrieval in enhancing AI response quality. This research serves as a critical reminder for developers in the AI/ML community about the limitations of static knowledge bases and the potential of integrating web search capabilities to improve model performance and reliability.