Puce Luca, Ceylan Halil İbrahim, Trompetto Carlo, Cotellessa Filippo, Schenone Cristina, Marinelli Lucio, Zmijewski Piotr, Bragazzi Nicola Luigi, Mori Laura
Department of Neuroscience, Rehabilitation, Ophthalmology, Genetics, Maternal and Child Health, University of Genoa, Italy.
Physical Education and Sports Teaching Department, Faculty of Kazim Karabekir Education, Atatürk University, Erzurum, Turkey.
Biol Sport. 2024 Oct;41(4):305-313. doi: 10.5114/biolsport.2024.141063. Epub 2024 Jul 23.
Nutrition is vital for athletic performance, especially in ultra-endurance sports, which pose unique nutritional challenges. Despite its importance, there exist gaps in the nutrition knowledge among athletes, and emerging digital tools could potentially bridge this gap. The ULTRA-Q, a sports nutrition questionnaire adapted for ultra-endurance athletes, was used to assess the nutritional knowledge of ChatGPT-3.5, ChatGPT-4, Google Bard, and Microsoft Copilot. Their performance was compared with experienced ultra-endurance athletes, registered sports nutritionists and dietitians, and the general population. ChatGPT-4 demonstrated the highest accuracy (93%), followed by Microsoft Copilot (92%), Bard (84%), and ChatGPT-3.5 (83%). The averaged AI model achieved an overall score of 88%, with the highest score in Body Composition (94%) and the lowest in Nutrients (84%). The averaged AI model outperformed the general population by 31% points and ultra-endurance athletes by 20% points in overall knowledge. The AI model exhibited superior knowledge in Fluids, outperforming registered dietitians by 49% points, the general population by 42% points, and ultra-endurance athletes by 32% points. In Body Composition, the AI model surpassed the general population by 31% points and ultraendurance athletes by 24% points. In Supplements, it outperformed registered dietitians by 58% points and the general population by 55% points. Finally, in Nutrients and in Recovery, it outperformed the general population only, by 24% and 29% points, respectively. AI models show high proficiency in sports nutrition knowledge, potentially serving as valuable tools for nutritional education and advice. AI-generated insights could be integrated with expert human judgment for effective athlete performance optimization.
营养对于运动表现至关重要，尤其是在超长耐力运动中，这类运动带来了独特的营养挑战。尽管其很重要，但运动员的营养知识存在差距，而新兴的数字工具可能会弥补这一差距。ULTRA-Q是一份针对超长耐力运动员改编的运动营养问卷，用于评估ChatGPT-3.5、ChatGPT-4、谷歌巴德和微软必应的营养知识。将它们的表现与经验丰富的超长耐力运动员、注册运动营养师和饮食学家以及普通人群进行了比较。ChatGPT-4的准确率最高（93%），其次是微软必应（92%）、巴德（84%）和ChatGPT-3.5（83%）。平均人工智能模型的总分为88%，其中身体成分方面得分最高（94%），营养素方面得分最低（84%）。在总体知识方面，平均人工智能模型比普通人群高出31个百分点，比超长耐力运动员高出20个百分点。人工智能模型在液体方面表现出卓越的知识，比注册饮食学家高出49个百分点，比普通人群高出42个百分点，比超长耐力运动员高出32个百分点。在身体成分方面，人工智能模型比普通人群高出31个百分点，比超长耐力运动员高出24个百分点。在补充剂方面，它比注册饮食学家高出58个百分点，比普通人群高出55个百分点。最后，在营养素和恢复方面，它仅比普通人群分别高出24个和29个百分点。人工智能模型在运动营养知识方面表现出很高的熟练度，有可能成为营养教育和建议的宝贵工具。人工智能生成的见解可以与专家的人类判断相结合，以有效地优化运动员的表现。
