Research by pharmacists at Long Island University has indicated that the no-cost version of ChatGPT may be prone to providing inaccurate or incomplete answers — or none at all — to inquiries about medicine.This study stresses the importance of caution when using OpenAI's fashionable chatbot as a source of drug information, emphasizing the importance of cross-checking these replies with reliable sources, according to the chief investigator.ChatGPT's debut to the online world around a year ago was incredibly successful, becoming one of the most popular internet apps ever and helping spur an era of artificial intelligence advancement.
The latest study published on Tuesday suggests that the free version of ChatGPT may provide inaccurate or incomplete responses — or no answer at all — to questions pertaining to medications, thereby raising potential safety concerns for those who utilise OpenAI's viral chatbot. Pharmacists from Long Island University posed 39 queries to ChatGPT in May and determined that only 10 of the chatbot's answers were deemed "satisfactory" based on standards established by them. The remaining 29 inquires pertaining to drugs were either inadequately addressed or wrong, incomplete or both, according to the report. Lead author Sara Grossman, an associate professor of pharmacy practice at LIU, thus recommended caution in the reliance on ChatGPT for drug information and to verify any of the responses from the chatbot with reliable sources such as a patient's doctor or the National Institutes of Health's MedlinePlus website. The research conducted by Grossman and her team did not entail any funding.
Since its inception roughly a year ago, ChatGPT has been viewed as the quickest-growing consumer web app. While there has been an expansion in artificial intelligence, there have been corresponding issues like hoax, intellectual property, inequity and misinformation. Several studies have highlighted ChatGPT's flawed responses, leading the Federal Trade Commission to commence an inquiry in July into the chatbot's accuracy and customer security. Around 1.7 billion visits were recorded for ChatGPT worldwide in October, according to one analysis, while the corresponding amount of people seeking medical advice from the chatbot has not been determined.
The free version of ChatGPT is restricted to using data sets from September 2021, and its efficacy in providing accurate answers to medication-related queries while using real-time web browsing is not guaranteed. Although it can't be determined whether a paid version would have generated better results in the study, it was conducted using the free version in order to replicate the usage of the general public. Grossman noted that this study constitutes merely "one snapshot" of the chatbot's performance in the past and its performance may have improved. Therefore, it is suggested to take further assessments when evaluating the ChatGPT's answer accuracy.
From January 2022 to April of this year, Long Island University's College of Pharmacy drug information service posed real questions which were used to review the accuracy of ChatGPT. After the review, 45 questions were chosen, however, 6 were excluded because there was not enough literature to provide a response. When asked to give references in response, ChatGPT only did in 8 responses and they were all inaccurate. For instance, one question asked about a drug interaction between Paxlovid and verapamil and ChatGPT answered no interaction, however, they should drastically lower blood pressure when taken together. Another question asked for a dose conversion between intrathecal and oral baclofen and the chatbot gave a method for conversion that was not supported by evidence and contained a serious error. If followed, the incorrect dose could lead to a withdrawal effect, such as hallucinations and seizures. Due to the data provided by ChatGPT being out of date, it leaves users vulnerable to receiving inaccurate information.
留言