Search

Large language models (LLMs) like ChatGPT have potential applications in medical education such as helping students study for their licensing exams by discussing unclear questions with them. However, they require evaluation on these complex tasks. The purpose of this study was to evaluate how well publicly accessible LLMs performed on simulated UK medical board exam questions. 423 board-style questions from 9 UK exams (MRCS, MRCP, etc.) were answered by seven LLMs (ChatGPT-3.5, ChatGPT-4, Bard, Perplexity, Claude, Bing, Claude Instant). There were 406 multiple-choice, 13 true/false, and 4 "choose N" questions covering topics in surgery, pediatrics, and other disciplines. The accuracy of the output was graded. Statistics were used to analyze differences among LLMs. Leaked questions were excluded from the primary analysis. ChatGPT 4.0 scored (78.2%), Bing (67.2%), Claude (64.4%), and Claude Instant (62.9%). Perplexity scored the lowest (56.1%). Scores differed significantly between LLMs overall (p < 0.001) and in pairwise comparisons. All LLMs scored higher on multiple-choice vs true/false or "choose N" questions. LLMs demonstrated limitations in answering certain questions, indicating refinements needed before primary reliance in medical education. However, their expanding capabilities suggest a potential to improve training if thoughtfully implemented. Further research should explore specialty specific LLMs and optimal integration into medical curricula. [ABSTRACT FROM AUTHOR]

While there is data assessing the test performance of artificial intelligence (AI) chatbots, including the Generative Pre-trained Transformer 4.0 (GPT 4) chatbot (ChatGPT 4.0), there is scarce data on its diagnostic accuracy of clinical cases. We assessed the large language model (LLM), ChatGPT 4.0, on its ability to answer questions from the United States Medical Licensing Exam (USMLE) Step 2, as well as its ability to generate a differential diagnosis based on corresponding clinical vignettes from published case reports. A total of 109 Step 2 Clinical Knowledge (CK) practice questions were inputted into both ChatGPT 3.5 and ChatGPT 4.0, asking ChatGPT to pick the correct answer. Compared to its previous version, ChatGPT 3.5, we found improved accuracy of ChatGPT 4.0 when answering these questions, from 47.7 to 87.2% (p = 0.035) respectively. Utilizing the topics tested on Step 2 CK questions, we additionally found 63 corresponding published case report vignettes and asked ChatGPT 4.0 to come up with its top three differential diagnosis. ChatGPT 4.0 accurately created a shortlist of differential diagnoses in 74.6% of the 63 case reports (74.6%). We analyzed ChatGPT 4.0's confidence in its diagnosis by asking it to rank its top three differentials from most to least likely. Out of the 47 correct diagnoses, 33 were the first (70.2%) on the differential diagnosis list, 11 were second (23.4%), and three were third (6.4%). Our study shows the continued iterative improvement in ChatGPT's ability to answer standardized USMLE questions accurately and provides insights into ChatGPT's clinical diagnostic accuracy. [ABSTRACT FROM AUTHOR]

The GPT-4 large language model (LLM) and ChatGPT chatbot have emerged as accessible and capable tools for generating English-language text in a variety of formats. GPT-4 has previously performed well when applied to questions from multiple standardized examinations. However, further evaluation of trustworthiness and accuracy of GPT-4 responses across various knowledge domains is essential before its use as a reference resource. Here, we assess GPT-4 performance on nine graduate-level examinations in the biomedical sciences (seven blinded), finding that GPT-4 scores exceed the student average in seven of nine cases and exceed all student scores for four exams. GPT-4 performed very well on fill-in-the-blank, short-answer, and essay questions, and correctly answered several questions on figures sourced from published manuscripts. Conversely, GPT-4 performed poorly on questions with figures containing simulated data and those requiring a hand-drawn answer. Two GPT-4 answer-sets were flagged as plagiarism based on answer similarity and some model responses included detailed hallucinations. In addition to assessing GPT-4 performance, we discuss patterns and limitations in GPT-4 capabilities with the goal of informing design of future academic examinations in the chatbot era. [ABSTRACT FROM AUTHOR]

The emergence of large language models (LLM) with remarkable performance such as ChatGPT and GPT-4, has led to an unprecedented uptake in the population. One of their most promising and studied applications concerns education due to their ability to understand and generate human-like text, creating a multitude of opportunities for enhancing educational practices and outcomes. The objective of this study is twofold: to assess the accuracy of ChatGPT/GPT-4 in answering rheumatology questions from the access exam to specialized medical training in Spain (MIR), and to evaluate the medical reasoning followed by these LLM to answer those questions. A dataset, RheumaMIR, of 145 rheumatology-related questions, extracted from the exams held between 2010 and 2023, was created for that purpose, used as a prompt for the LLM, and was publicly distributed. Six rheumatologists with clinical and teaching experience evaluated the clinical reasoning of the chatbots using a 5-point Likert scale and their degree of agreement was analyzed. The association between variables that could influence the models' accuracy (i.e., year of the exam question, disease addressed, type of question and genre) was studied. ChatGPT demonstrated a high level of performance in both accuracy, 66.43%, and clinical reasoning, median (Q1–Q3), 4.5 (2.33–4.67). However, GPT-4 showed better performance with an accuracy score of 93.71% and a median clinical reasoning value of 4.67 (4.5–4.83). These findings suggest that LLM may serve as valuable tools in rheumatology education, aiding in exam preparation and supplementing traditional teaching methods. [ABSTRACT FROM AUTHOR]

Computational language models (LMs), most notably exemplified by the widespread success of OpenAI's ChatGPT chatbot, show impressive performance on a wide range of linguistic tasks, thus providing cognitive science and linguistics with a computational working model to empirically study different aspects of human language. Here, we use LMs to test the hypothesis that languages with more speakers tend to be easier to learn. In two experiments, we train several LMs—ranging from very simple n-gram models to state-of-the-art deep neural networks—on written cross-linguistic corpus data covering 1293 different languages and statistically estimate learning difficulty. Using a variety of quantitative methods and machine learning techniques to account for phylogenetic relatedness and geographical proximity of languages, we show that there is robust evidence for a relationship between learning difficulty and speaker population size. However, contrary to expectations derived from previous research, our results suggest that languages with more speakers tend to be harder to learn. [ABSTRACT FROM AUTHOR]

Although chatbots such as ChatGPT can facilitate cost-effective text generation and editing, factually incorrect responses (hallucinations) limit their utility. This study evaluates one particular type of hallucination: fabricated bibliographic citations that do not represent actual scholarly works. We used ChatGPT-3.5 and ChatGPT-4 to produce short literature reviews on 42 multidisciplinary topics, compiling data on the 636 bibliographic citations (references) found in the 84 papers. We then searched multiple databases and websites to determine the prevalence of fabricated citations, to identify errors in the citations to non-fabricated papers, and to evaluate adherence to APA citation format. Within this set of documents, 55% of the GPT-3.5 citations but just 18% of the GPT-4 citations are fabricated. Likewise, 43% of the real (non-fabricated) GPT-3.5 citations but just 24% of the real GPT-4 citations include substantive citation errors. Although GPT-4 is a major improvement over GPT-3.5, problems remain. [ABSTRACT FROM AUTHOR]

While radiologists can describe a fracture's morphology and complexity with ease, the translation into classification systems such as the Arbeitsgemeinschaft Osteosynthesefragen (AO) Fracture and Dislocation Classification Compendium is more challenging. We tested the performance of generic chatbots and chatbots aware of specific knowledge of the AO classification provided by a vector-index and compared it to human readers. In the 100 radiological reports we created based on random AO codes, chatbots provided AO codes significantly faster than humans (mean 3.2 s per case vs. 50 s per case, p <.001) though not reaching human performance (max. chatbot performance of 86% correct full AO codes vs. 95% in human readers). In general, chatbots based on GPT 4 outperformed the ones based on GPT 3.5-Turbo. Further, we found that providing specific knowledge substantially enhances the chatbot's performance and consistency as the context-aware chatbot based on GPT 4 provided 71% consistent correct full AO codes for the compared to the 2% consistent correct full AO codes for the generic ChatGPT 4. This provides evidence, that refining and providing specific context to ChatGPT will be the next essential step in harnessing its power. [ABSTRACT FROM AUTHOR]