Shortcomings of ChatGPT

Abstract

There appears to be little to no discussion of the major shortcomings of ChatGPT that are relevant to chemical educators, students, and researchers. The performance of ChatGPT, described in three recent papers in J. Chem. Educ., was critically examined along with its performance on assignments in a 100-level general education chemistry course for nonscience majors. ChatGPT's possible role as an assistant in searching the literature relevant to the biogeochemistry of arsenic was also examined. ChatGPT cannot perform mathematical operations reliably, makes conceptual errors, and fabricates plausiblelooking citations, the purported contents of which are only partly accurate.

ChatGPT Needs a Chemistry Tutor Too

“…Although ChatGPT is a language-based model that may struggle with complex chemical concepts such as structural features, formulas, and mathematical aspects, its investigation as a potential learning tool in a fully AI-driven environment remains valuable. Additionally, the recent studies have revealed that ChatGPT has two major limitations: it struggles with problems that require mathematical analysis, and it can fabricate references and contents. , The present study involved ChatGPT as a support tool for learners in an introductory chemistry course by examining its outputs, assessing the accuracy of its answers, and understanding how it processed prompts. The Introductory Chemistry course prepares learners with the fundamental knowledge and abilities needed to excel in the first semester of General Chemistry I.…”

An Analysis of AI-Generated Laboratory Reports across the Chemistry Curriculum and Student Perceptions of ChatGPT

“…Additionally, Emenike and Emenike summarized a variety of ways that ChatGPT could be used to serve students, faculty, and administrators for teaching and learning, research, and professional activities . Even more recent reports have captured some of AI’s shortcomings for chemical education and used ChatGPT responses as focal points for student critiques in a problem solving exercise . One comparison for ChatGPT may be Wikipedia, which debuted in 2001 and also vastly increased the availability of information; opinions of Wikipedia were initially polarized, but it has now been embraced for the learning opportunities it provides to students. − …”

Comparing Student and Generative Artificial Intelligence Chatbot Responses to Organic Chemistry Writing-to-Learn Assignments

“…While we provide a baseline comparison between chatbot and student responses on features of mechanistic reasoning, there may be additional insight to be gleaned from examining the influence of follow-up questions. Additionally, initial reports suggest that chatbots may fabricate information, exhibit explanatory biases, or demonstrate flawed reasoning; , when generating responses for this study, we identified both similar and additional trends in chatbot responses. For example, it was apparent that chatbot responses may have been more sophisticated in some ways than what we might expect from students, despite being less detailed with respect to the mechanistic reasoning features identified by the ML models.…”

“…Attention in the chemistry education literature has turned toward understanding the role generative AI may have for the future of assessments in chemistry . Initial reports within the chemistry education literature indicate that generative AI responses tend to exhibit mixed accuracy with respect to factual or knowledge-based questions, especially facing challenges with problems requiring analysis or interpretation; findings within these studies focus on the correctness of chemistry content, suggesting that generative AI responses are generally well-written with persuasive logic and reasoning. ,,− The present study contributes to this emerging literature by focusing specifically on evaluating the capabilities of generative AI chatbots to engage in mechanistic reasoning on organic chemistry WTL assignments. This study focuses on responses from two AI chatbots: ChatGPT and Bard.…”