Answers are grounded in this paper and cite their sources.
Since the release of ChatGPT (OpenAI, San Francisco, CA) in November 2022, several additional organizations have released large language models (LLMs), including Google (PaLM; Alphabet Inc., Mountain View, CA) and the Technology Innovation Institute (Falcon; Abu Dhabi, United Arab Emirates). Most recently, on November 4, 2023, Elon Musk's xAI (San Francisco, CA) announced the release of a novel artificial intelligence (AI) model, Grok. According to the team, Grok is different from some current LLMs because the system is able to synthesize responses to user commands with “real-time knowledge” from the X platform (previously known as Twitter; San Francisco, CA), allowing users to have access to up-to-date information.1 Additionally, Grok has been labeled by itself as a “powerful research assistant for anyone, helping them to quickly access relevant information, process data, and come up with new ideas” and “intended to answer almost anything and . . . even suggest what questions to ask!”1 Grok is powered by Grok-1, which was trained by Grok-0 with 33 billion parameters.1 According to xAI, Grok surpassed other models, including ChatGPT 3.5, but fell short of GPT-4. Given the novelty of Grok, the authors of this study wanted to explore the potential of this artificial intelligence (AI) model for aesthetic plastic surgery by applying Grok to the lead author's initial study addressing utilization of ChatGPT for the creation of novel systematic review topics in plastic surgery.2,3 We believe this is the first study that applies Grok to plastic surgery. To encompass the diverse practice of aesthetic plastic surgery, Grok was asked to give “novel systematic review ideas” on certain topics within “aesthetic plastic surgery” (Figure 1). Grok was accessed through a paid X account membership. The AI model produced a total of 20 novel systematic review ideas for 12 topics within aesthetic plastic surgery. Within each topic, the model was commanded to produce 10 general and 10 specific review ideas related to that topic (Table 1). In sum, Grok produced a total of 240 topics. To determine the accuracy of Grok, a literature search and review was conducted in PubMed (National Institutes of Health, Bethesda, MD); EMBASE (Elsevier, Amsterdam, the Netherlands); CINAHL (EBSCO Information Services, Birmingham, AL); and Cochrane Library (Cochrane, London, UK). Example of command given to Grok regarding 10 general systematic review topics on blepharoplasty. Systematic Review Ideas on Blepharoplasty Generated by AI Grok Systematic Review Ideas on Blepharoplasty Generated by AI Grok We determined that Grok had an overall accuracy of proposing novel systematic review topics of 74.2% (Table 2). When stratified by general and specific topics, Grok had an accuracy of 61.7% for general systematic review ideas and 86.7% for specific systematic review ideas. The highest scoring categories were skin tightening, skin treatment (chemical peels and hydrafacials), and breast implant removal (19, 18, and 17 novel topics respectively). The lowest scoring categories were breast augmentation, dermal fillers, and abdominoplasty (10, 10, and 13 novel topics respectively) (Table 3). Overall Number of Novel Systematic Review Topics Generated by AI Grok Overall Number of Novel Systematic Review Topics Generated by AI Grok Number of Novel AI Grok–Generated Topics in Operative and Nonoperative Aesthetic Surgery Number of Novel AI Grok–Generated Topics in Operative and Nonoperative Aesthetic Surgery To better understand how Grok compared to some of the alternative LLMs, we compared our data to our previously published assessment of the accuracy of ChatGPT and GPT-4 for systematic review creation.4 To our surprise, we found that Grok performed better in all categories when compared to ChatGPT and GPT-4, including overall accuracy as well as accuracy when stratified by general or specific topics (ChatGPT 55% overall, 35% general topics, and 75% specific topics; GPT-4 65% overall, 50% general topics, and 80% specific topics). One of the key features of Grok is its connection to the X platform, allowing the AI model to have access to a wealth of real-time information, which may make it more accurate and better suited for some users. Also, as Grok relies on the X platform, this may mean that this LLM is utilizing a greater amount of human-generated content than other AI models. In addition, Grok has demonstrated better performance than ChatGPT in mathematical problem-solving ability (MATH), massive multitask language understanding (MMLU), and HumanEval, all 3 of which are machine-learning benchmarks that are able to measure mathematic and reasoning capabilities of AI systems; these may help evaluate the strengths and limitations of the model.1 However, it is important to note that ChatGPT has a broader range of capabilities, making it more convenient and suitable for general tasks. Grok has several limitations, including the generation of false or contradictory information, a low parameter count, a rushed development leading to training limitations, and dependence on X data.1 Furthermore, given this information, the choice of whether to utilize ChatGPT or Grok may be user dependent, with Grok potentially being more helpful to users seeking to access real-time information with a focused approach. All in all, within the past year there has been a surge in the development of large language models. Our initial research on ChatGPT and its potential effects for plastic surgery demonstrated promising applications for various aspects of our field, including research, patient care, and resident education. With the introduction of Grok, the authors of this study believe that large language models are a step closer to being an effective aid, particularly for research enhancement and resident education. Furthermore, by integrating LLMs such as Grok into their practice, plastic surgeons may be able to improve patient outcomes, streamline their workflow, and aid in the promotion of research-based education. The authors declared no potential conflicts of interest with respect to the research, authorship, and publication of this article. The authors received no financial support for the research, authorship, and publication of this article.
No takes yet. Share an insight, caveat, or question.
Gupta et al. (2024) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: