Assessing the accuracy of large language models in extracting latest cricket information
Indika Wickramasinghe · Scientific Journal of Sport and Performance · 2025
The development of large language models (LLMs) is making waves across various fields, bringing numerous benefits and innovations. At the same time, cricket is growing rapidly in popularity worldwide. Given this context, it's a great moment to explore how well LLMs can keep up with the latest cricket knowledge. This study evaluates the performance of three LLMs Co-Pilot, ChatGPT, and Liner in generating accurate summaries of bilateral Test and One Day Internationals (ODI) cricket series played in 2024. The evaluation focused on three main tasks: reporting series results, identifying the top three batsmen with their scores, and listing the top three bowlers with their wickets. Among the models, Co-Pilot stood out, consistently delivering the highest accuracy across all tasks and formats, especially for matches involving Australia, India, and South Africa. ChatGPT showed mixed results, excelling in some areas but struggling with task-specific accuracy. Liner, on the other hand, had the lowest accuracy and faced significant challenges in providing relevant detailed cricket-related information. The study also noted instances where the models generated unrelated or incorrect outputs, highlighting the need to validate LLM-generated cricket data to ensure it is reliable and correct.