BELL: Benchmarking the Explainability of Large Language Models

Large language models have revolutionized natural language processing and generative Artificial Intelligence (AI), as shown by numerous foundational studies [1]. These models ' exceptional capabilities have attracted significant attention, enabling a wide range of applications. LLMs are utilized for tasks such as translation [2], content generation, content summarization, article writing [3], as well as enhancing search function s (Bing Chat [4]) etc., The impact of LLMs extends to fields like software develo pment, with models like Code Llama [5] aiding engineers . Their applications also span finance sector [6], scientific research [7] [8], including areas such as arts [9], education [10], oceanography [11], law [12], political science [13], medicine [14] [15], showcasing their broad and diverse influence. However, t he exponential rise in use of LLMs also brings challenges related to their explainability and interpretability.