Statistical issues in general (biostatisticians meet AI)
December 17, 2025 A previously published paper, ChatGPT as a Tool for Biostatisticians: A Tutorial on Applications, Opportunities, and Limitations by Dobler et al. that also appeared in Statistics in Medicine, which according to Zhu provides a timely and insightful assessment of how large language models (LLMs) can assist in biostatistical work. The authors had showcased several real-world use cases to illustrate ChatGPT's capabilities and downsides in the biostatistical domain. Zhu states that Dobler et al. had set a pragmatic tone and reinforced that LLMs, while transformative, are no substitute for sound statistical judgment. Zhu then also states that they applaud the authors for this balanced, real-world tutorial, which serves as a starting point for biostatisticians interested in harnessing artificial intelligence (AI) in practice. The author mentions that these LLMs should be given qualifying exams to see if they can pass since they are becoming more adapt and analyzing, synthesizing, and interpreting data, like a human. This could be used as a benchmark for biostatistics to see how well the LLMs function in certain biostatistical areas. As Zhu pointed out, the rise of LLMs will demands more statistical rigor and critical thinking from us, not less. The first reason is that LLMs are not infallible as shown by Dobler et al and another paper. Although the LLM can produce output that appears perfectly fluid and convincing, it could actually be subtly wrong or based on faulty reasoning. The second reason is that effective use of LLMs requires critical thinking in the guidance that biostatisticians would provide. LLMs often need precisely crafted prompts and follow-up questions to produce correct outputs. The aforementioned tutorial paper demonstrates this well that while ChatGPT-4o could eventually solve some biostatistics problems, it could only do this with careful guidance and multiple attempts. Essentially, a biostatistician must know when the