Large language models (LLMs) have become extensively used among users across diverse settings. Yet, with the complex nature of these large-scale artificial intelligence (AI) systems, leveraging their capabilities effectively is yet to be explored. In this study, we looked at the types of communication errors that occur in interactions between humans and ChatGPT-3.5 in Arabic. A corpus of six Arabic-language consultations was collected from an online mental health support forum. For each consultation, the researchers provided the user’s Arabic queries to ChatGPT-3.5 and analyzed the system’s responses. The study identified 102 communication errors, mostly grammatical and repetitions. Other errors involved contradictions, ambiguous language, ignoring questions, and lacking sociality. By examining the patterns and types of communication errors observed in ChatGPT’s responses, the study is expected to provide insights into the challenges and limitations of current conversational AI systems, particularly in the context of sensitive domains like mental health support.
Al-Otaibi et al. (Mon,) studied this question.
Synapse has enriched 5 closely related papers on similar clinical questions. Consider them for comparative context: