ChatGPT: Unraveling User Challenges & Proposing Targeted Improvements

Abstract

This research paper investigates the multifaceted challenges users encounter with chatbots, focusing specifically on the widely adopted Generative Pre-trained Transformer (GPT) models. Despite GPT's commendable proficiency in natural language understanding, issues arise, notably in the realm of mathematical calculations. A comprehensive literature review outlines the strengths and limitations of GPT across diverse user queries. Additionally, the paper sheds light on user-reported concerns, including GPT's neutral responses that may not address queries adequately, occasional unreliability, and security vulnerabilities. The study meticulously assesses GPT's performance in mathematical problem-solving, uncovering instances of inaccuracies and user dissatisfaction. The discussion delves into potential contributing factors, such as model architecture and training data, while proposing enhancements to mitigate these challenges. Moreover, the research acknowledges broader concerns, such as the delivery of excessively long answers and the lack of artistic touch and creativity in responses. Noteworthy is the observation that identical questions often yield remarkably similar responses, indicating a potential limitation in the model's diversity and adaptability. The paper concludes by underscoring the importance of addressing these challenges for an enhanced user experience and recommends future research directions for refining the ability of language models like GPT.

Ähnliche Arbeiten