In last month’s newsletter, we discussed the capabilities of ChatGPT and demonstrated an example of its output. Now, several months since its public release late last year, we will delve into the excitement surrounding its applications and problems that have been identified with its use, and how it or its future iterations could change the future for all of us.
A comment published in nature exemplifies some of the feelings that researchers have regarding ChatGPT, perhaps best described as a sort of cautious optimism mixed with feelings of looming concern. The authors note that ChatGPT has already been used to draft papers and summarize literature, among other applications. While they recognize the excitement ChatGPT brings, they also advise caution due to problems such as those regarding accountability, accuracy, and the blackbox nature of the language model. ChatGPT’s well-known ability to sound convincing makes it all the more dangerous due to its known tendency to occasionally produce biased answers, factually inaccurate statements, or even logically incoherent ones (see this GitHub repository of ChatGPT failures). As the model is not open-source and even its creators have limited insight into its exact workings, there is genuine concern over the damage that could be caused by reliance on such an imperfect, but otherwise convincing, facsimile of human expertise.
Despite these concerns, ChatGPT remains a symbol of what lies next on the horizon. Researchers have already produced surprising amounts of research directly relating to the language model, including this pre-print regarding the use of ChatGPT to simplify radiology reports, potentially allowing increased patient autonomy through enabling patients to understand information previously too opaque or technical for laypeople. Yet another published article discusses the using ChatGPT to assist in the generation of literature reviews, which is traditionally a laborious process.
What is clear is that ChatGPT already carries frightening potential for disruption in many fields. But, perhaps even more capable of inspiring both terror and hope, is the prospect of what might come next. This language model in the infancy of the age of AI is already capable of so much. We might not even be able to imagine what advances will be made another ten years from now.
Click here for the Japanese version.