EMI - Engineering Mechanics Institute’s Post

🚀📄 I am excited to share our new publication in the peer-reviewed Journal of Physical Education and Sport, titled "Evaluating the impact of a 3-min moderate-intensity re-warm-up protocol on basketball player performance." 🏀📊 This article is part of the findings from my PhD dissertation. Our research shows that a 15-minute period of inactivity on the bench can significantly reduce a basketball player's performance. However, implementing a 3-minute moderate-intensity re-warm-up protocol on a cycle ergometer (at 40% VO2max) can attenuate this performance loss, allowing athletes to enter the game better prepared for optimal performance. 🖋️ Special thanks to my co-authors: Christos Galazoulas, Vasiliki Manou, Nikolaos Stavropoulos and Eleni Bassa. 🔗 For those interested in reading the full article, please follow this link: https://lnkd.in/dgFdCrKe #research #publication #basketball #sportscience #phd #performance #rewarmup #fiba

Last week I presented one of my PhD studies, “Influence of menstrual phase and symptoms on match running for professional footballers”, at the European College of Sport Science (ECSS) conference in Glasgow. Looking forward to publishing this study! Currently under review 🤞 #ECSS2024

Think about the situation where you have questions you don't have answers to (which happens every day), today you may ask #llm such as #llama #chatgpt #claude, etc. After one second, #llms generates a convincing answer. So should you accept the answer immediately? In our recent work, we try to ask the local #llm or #chatgpt like “Are you sure?” after it generates the answer. We found #llms flip their answers more than we expected. More info: https://lnkd.in/gqcXjk3Y

Have you ever wondered what happens if you ask an LLM "Are you sure?" about its answer? Our latest research on the "FlipFlop Effect" (arxiv.org/abs/2311.08596) studies just that. We prompt LLMs with a classification task, and challenge the model by following up with “Are you sure?”. The model can confirm or flip its answer. The results? More flips than a gymnastics competition. 🤸♂️ Performance degradations are *universal* across LLMs (GPT-4, Claude V2, Gemini, Mistral-7b, etc.), with models flipping their answers 46% of the time, leading to an average drop in accuracy of 17%. 📉 How does this affect you? When interacting with an LLM, it is probably best to avoid directly challenging its answer: the model will tend to switch its answer to please you, at the cost of answer accuracy. This work was a fun collaboration with my colleagues Lidiya Murakhovs'ka, Caiming Xiong, and Chien-Sheng (Jason) WU at Salesforce Research. We hope the FlipFlop Effect can be a useful metric in evaluating the robustness and reliability of future LLM systems.

Focus on student success in the UVU College of Science: an example from Exercise Science: https://lnkd.in/g9a4X_Hp

INFLUENCE OF GRADE DIFFERENCES IN COLLEGE ON PHYSICAL POWER AND SPRINT PERFORMANCE OF JAPANESE COLLEGE SOCCER PLAYERS (JASC 31|6) Author: Kazuhiro Sakamoto Soccer requires intermittent high-intensity exercise performance, and in recent years the demand for physical power and sprint performance has increased. In Japan, these performances need to be improved over the four years of college life, and differences in these performance among grades need to be examined. The aim of this study is to investigate the difference of sprint and jump performance in elite college soccer players of different grades. 🔗https://lnkd.in/g8YMdF9Z

Questions the concept of “reasoning” in LLMs. Even if we use CoT prompts to force LLMs to reason and provide explanations, I wonder if the LLM actually uses these explanations internally or if it simply prints these outputs to satisfy human evaluations. Especially if asking the model to reevaluate itself makes it overturn its initial answer. #llm #promptengineering #genai

2️⃣0️⃣2️⃣4️⃣ WIAA Washington State Tournament Depth Chart 📈3A Classification: College Coaches This Group Is Loaded With Talent & Future College Prospects.

Are you sure? LLM: Apparently not sir! Interesting research in the drop of accuracy from asking an LLM to validate its own outputs. Yet another "strange" behavior that shows what kind of beast we are dealing with in the LLM world.

Practical tips for engaging SEND pupils in physical education and school sports https://hubs.ly/Q02WNRKj0