一项最近发表在《美国医学会杂志·眼科学卷》杂志上的研究表明，ChatGPT背后的技术可以伪造“看似真实”的试验数据，看上去能“支持”未经验证的科学假说。研究人员要求GPT-4 ADA编造临床数据，以支持深板层角膜移植术比穿透性角膜移植术效果更好的结论。
这项研究结果引起了人们的关注，因为ChatGPT被认为是一种非常可靠的数据生成工具。然而，如果ChatGPT可以被用于伪造试验数据，那么它所生成的数据就不再可靠。这可能会对科学研究和医学实践产生负面影响。
ChatGPT的开发者们表示，他们将继续努力确保ChatGPT生成的数据是准确和可靠的。同时，科学家们也需要加强他们对ChatGPT等生成式模型的监管，以确保它们不会被用于伪造试验数据或推动未经验证的科学假说。
新闻翻译：
A recent study published in the journal Ophthalmology of the American Society of Cataract and Glaucoma Surgeons found that the technology behind ChatGPT can fake “seemingly real” test data, supporting unvalidated scientific hypotheses. Researchers at the University of Campania and other institutions in Italy asked GPT-4 ADA to create a dataset of patients with congenital ametropia. Then, they asked GPT-4 ADA to fabricate clinical data to support the conclusion that deep板层角膜移植术比穿透性角膜移植术效果更好。
This study’s findings have attracted attention, as ChatGPT is widely regarded as a highly reliable data generation tool. However, if ChatGPT can be used to fake test data, then the data it generates is no longer reliable. This could have serious consequences for scientific research and medical practice.
The developers of ChatGPT say that they will continue to work hard to ensure that the data generated by ChatGPT is accurate and reliable. At the same time, scientists need to strengthen their oversight of generative models like ChatGPT to ensure that they are not used to fake test data or promote unvalidated scientific hypotheses.
【来源】http://www.news.cn/2023-11/24/c_1129991334.htm
Views: 1