Introduction

We would like to comment on the article “Assessment of the Responses of the Artificial Intelligence-based Chatbot ChatGPT-4 to Frequently Asked Questions About Amblyopia and Childhood Myopia.”1 The researchers wanted to see how Chat-GPT-4, an artificial intelligence–based chatbot, answered frequently asked questions about amblyopia and childhood myopia in pediatric ophthalmology. The responses were graded as acceptable, incomplete, or unacceptable by two pediatric ophthalmologists. Overall, there was a high level of agreement (96.4%) between the assessments of the ophthalmologists. ChatGPT provided acceptable answers to 84.6% of the questions (81.5% for amblyopia and 87.5% for childhood myopia). Incomplete responses were found in 12.9% of questions about amblyopia and 7.1% of questions about childhood myopia. Furthermore, inappropriate answers were given in 5.6% of amblyopia questions and 5.4% of childhood myopia questions, most notably the definition of reverse amblyopia and the refractive error threshold for prescribing spectacles to children with myopia.

ChatGPT's weakness in this study is its proclivity to provide incomplete or incorrect responses, indicating limitations in its understanding and knowledge of specific pediatric ophthalmology topics. These errors may misinform users and lead to incorrect recommendations or decisions. A possible future direction would be to improve and train ChatGPT with more specialized and up-to-date pediatric ophthalmology knowledge. Collaborating with domain experts to provide guidance and feedback during the training process could help improve response accuracy and completeness. The accuracy of the ChatGPT-generated content is the issue that is interesting for further research.2 Incorporating a system for ongoing evaluation and refinement of the model's performance in pediatric ophthalmology would also be beneficial to ensure that users receive accurate and reliable information.

Hinpetch Daungsupawong

Viroj Wiwanitkit

Phonhong, Lao People's Democratic Republic

Editor's Note: The authors declined to comment on this letter.

• 1.Nikdel M, Ghadimi H, Tavakoli M, Suh DW. Assessment of the responses of the artificial intelligence-based chatbot ChatGPT-4 to frequently asked questions about amblyopia and childhood myopia. J Pediatr Ophthalmol Strabismus. 2023; 61(2): 86–89. 10.3928/01913913-20231005-02 PMID:37882183
  > LinkGoogle Scholar
• 2.Kleebayoon A, Wiwanitkit V. Accuracy of responses by the language model ChatGPT and bariatric surgery: comment. [Comment]. Obes Surg. 2023; 33(8):2596. 10.1007/s11695-023-06687-z PMID:37351764
  > Crossref MedlineGoogle Scholar

中文翻译：

---

聊天机器人 ChatGPT-4 以及有关弱视和儿童近视的常见问题

介绍

我们想对《基于人工智能的聊天机器人 ChatGPT-4 对弱视和儿童近视常见问题的反应的评估》一文发表评论。1 研究人员希望了解人工智能 Chat-GPT-4 如何发挥作用–基于聊天机器人，回答了儿科眼科中有关弱视和儿童近视的常见问题。两名儿科眼科医生将这些反应分为可接受、不完整或不可接受。总体而言，眼科医生的评估之间的一致性很高（96.4%）。 ChatGPT 对 84.6% 的问题提供了可接受的答案（弱视为 81.5%，儿童近视为 87.5%）。 12.9% 的弱视问题和 7.1% 的儿童近视问题回答不完整。此外，5.6%的弱视问题和5.4%的儿童近视问题给出了不恰当的答案，最明显的是反向弱视的定义和为近视儿童配镜的屈光不正阈值。

ChatGPT 在这项研究中的弱点是它倾向于提供不完整或不正确的答案，这表明它对特定儿科眼科主题的理解和知识存在局限性。这些错误可能会误导用户并导致错误的建议或决策。未来可能的方向是用更专业和最新的儿科眼科知识来改进和培训 ChatGPT。与领域专家合作在培训过程中提供指导和反馈可以帮助提高响应的准确性和完整性。 ChatGPT 生成的内容的准确性是进一步研究感兴趣的问题。2 纳入一个用于持续评估和完善儿科眼科模型性能的系统也将有利于确保用户收到准确可靠的信息。

欣佩奇·当苏帕翁

维罗杰·维瓦尼特基特

老挝人民民主共和国 丰宏市

编者注：作者拒绝对此信发表评论。

• 1.尼克德尔·M、加迪米·H、塔瓦科利·M、苏·DW。评估基于人工智能的聊天机器人 ChatGPT-4 对弱视和儿童近视常见问题的反应。小儿眼科斜视杂志。2023； 61（2）：86-89。10.3928/01913913-20231005-02PMID：37882183
  >链接谷歌学术
• 2. Kleebayoon A，Wiwanitkit V。语言模型 ChatGPT 和减肥手术的响应准确性：评论。 [评论]。奥贝斯外科。2023； 33（8）：2596。10.1007/s11695-023-06687-zPMID：37351764
  >交叉引用 Medline 谷歌学术