Everything we thought about running injury development was wrong, Danish study shows

Study with over 5,200 runners shows that running injuries do not develop gradually over time – but most often occur during a single running session.

Millions of runners are therefore receiving incorrect guidance from sports watches, researcher warns.

Aarhus University

Facebook

XLinkedInWeChatBlueskyMessageWhatsAppEmail

A new study from Aarhus University turns our understanding of how running injuries occur upside down.

The research project, which is the largest of its kind ever conducted, shows that running-related overuse injuries do not develop gradually over time, as previously assumed, but rather suddenly – often during a single training session.

"Our study marks a paradigm shift in understanding the causes of running-related overuse injuries. We previously believed that injuries develop gradually over time, but it turns out that many injuries occur because runners make training errors in a single training session," explains Associate Professor Rasmus Ø. Nielsen from the Department of Public Health at Aarhus University, who is the lead author of the study.

The study followed 5,205 runners from 87 countries over 18 months and shows that injury risk increases exponentially when runners increase their distance in a single training session compared to their longest run in the past 30 days. The longer the run becomes, the higher the injury risk.

Incorrect guidance for millions of runners

According to Rasmus Ø. Nielsen, the results cast critical light on how the tech industry has implemented so-called "evidence." Millions of sports watches worldwide are equipped with software that guides runners about their training – both for training optimization and injury prevention.

However, the algorithm used for injury prevention is built on very thin scientific grounds, according to Rasmus Ø. Nielsen.

"This concretely means that millions of runners receive incorrect guidance from their sports watches every day. They think they are following a scientific method to avoid injuries, but in reality they are using an algorithm that cannot predict injury risk at all," he says.

 

 

Non-existing evidence behind guidance

The current algorithm, called "Acute:Chronic Workload Ratio" (ACWR), was introduced in 2016 and is now implemented in equipment from companies that produce sports watches, while organizations and clinicians, such as physiotherapists, also use the algorithm.

The ACWR algorithm calculates the ratio between acute load (last week's training) and chronic load (average of the past 3 weeks). The algorithm recommends a maximum 20% increase in training load to minimize injury risk.

According to Rasmus Ø. Nielsen, the algorithm was originally developed for team sports and was based on a study with 28 participants. Due to the few participants in the study combined with data manipulation, the evidence base for using the algorithm to prevent running injuries is therefore "non-existent."

Realtime guidance

The research team has therefore worked for the past eight years to develop a new algorithm that will be much better at preventing injuries for runners.

Rasmus Ø. Nielsen emphasizes that he and the other researchers behind the study have no commercial interests in launching a new algorithm as a potential replacement for a method he himself criticizes.

The algorithm will be made freely available to runners, companies, clinicians and organizations who can use it actively to guide training and injury prevention.

Rasmus Ø. Nielsen hopes that the new insights will be implemented in existing technology.

"I imagine, for example, that sports watches with our algorithm will be able to guide runners in real-time during a run and give an alarm if they run a distance where injury risk is high. Like a traffic light that gives green light if injury risk is low; yellow light if injury risk increases and red light when injury risk becomes high," explains Rasmus Ø. Nielsen.

 

Facts about the study

The Garmin-Runsafe Running Health Study followed 5,205 runners from 87 countries over 18 months.

Participants recorded 588,071 running sessions, and 35 percent of participants sustained a running-related injury during the study.

The study documents concrete risk increases with increased running distances compared to the longest run in the past 30 days: 

10-30 percent increase: 64 percent increased injury risk

30-100 percent increase: 52 percent increased injury risk

Over 100 percent increase: 128 percent increased injury risk

The study also shows that injury risk increases with progressions over 1 percent (in the interval between 1 and 10 percent), which questions whether the frequently used 10% rule for safe training progression is accurate.

 

About the research results

Study type: Cohort study with 18 months follow-up period

Collaborators: Garmin International (participant recruitment) and researchers from Denmark, Sweden, Luxembourg and Australia.

External funding: The study received external funding from Aarhus University Research Foundation and the Danish Rheumatism Association. Garmin and external funders had no influence on study design, research questions, data collection, data processing and statistical analysis and/or interpretation of data as well as the writing and publication process.

Conflict of interest: None

Link to scientific article: How much running is too much? Identifying high-risk running sessions in a 5200-person cohort study | British Journal of Sports Medicine

---

Journal

British Journal of Sports Medicine

Method of Research

Observational study

Subject of Research

People

Article Title

How much running is too much? Identifying high-risk running sessions in a 5200-person cohort study

Article Publication Date

7-Jul-2025

Plate tectonics – Mineral olivine is crucial for heat transport in the mantle

University of Potsdam

Facebook

XLinkedInWeChatBlueskyMessageWhatsAppEmail

 

image: 

Sketch of subducting slab and transport of water-bearing minerals to the Mantle Transition Zone

view more 

Credit: Enrico Marzotto

Due to the radiative thermal conductivity of the mineral olivine, only oceanic plates over 60 million years old and subducting at more than 10 centimeters per year remain sufficiently cold to transport water into the Earth's deep mantle. This was found by scientists from the University of Potsdam and from the Helmholtz Centre for Geosciences (GFZ) Potsdam, together with international colleagues, by measuring the transparency of olivine under conditions in Earth’s mantle for the first time. Their results were published in the journal “Nature Communications”.

The outer layer of our planet, the lithosphere, is fragmented into several rigid plates, which are drifting on top of the warm and relatively soft mantle. The collision between plates causes the heavier plate to sink into the Earth's mantle. This sinking process takes the name of ‘subduction’, while the subducting plate is called ‘slab’. Usually, the oceanic lithosphere is heavier than the continental one, because it is made of denser minerals like olivine, which accounts for 80 percent of the oceanic lithosphere. Olivine is also the predominant mineral in the Earth's outer shell, representing 60% of the upper mantle (40-410 km of depth). While subducting, the cold slabs are progressively heated by the warm ambient mantle through heat diffusion, a process that involves heat conduction and heat radiation. Understanding slab heating processes is fundamental to explaining the occurrence of deep earthquakes, and the presence of water at more than 600 km of depth.

“We measured, for the first time, the transparency of olivine inside the Earth,” says geodynamicist Enrico Marzotto from the Institute of Geosciences of the University of Potsdam. “These measurements demonstrate that olivine is infrared transparent even at the extreme pressure and temperature conditions of Earth’s interior.” The heat transport by radiation accounts for approximately 40 percent of the total heat diffused in the olivine-rich upper mantle. Therefore, radiative thermal conductivity plays an important role in slab heating and can have far-reaching effects on the density and the rigidity of the subducting plates, and their capacity to carry water into Earth’s interior.

With two-dimensional slab thermal evolution models the team could show that the rapid heating enhanced by radiative heat transport induces the breakdown of water-bearing minerals at shallower depth. “This could potentially explain the occurrence of earthquakes in the slab at more than 70 kilometers of depth,” says Marzotto. “Consequently, only slabs that are more than 60 Million years old and sinking faster than 10 centimeters per year, remain sufficiently cold to transport the water-bearing minerals down to the Mantle Transition Zone (MTZ) in 410 to 660 kilometers depth.” The MTZ is considered the largest water reservoir on our planet, potentially containing up to three times more water than the Earth's oceans.

“Our study also provides numerical tools to compute the lifetime of thermal anomalies in the mantle and their geodynamic behavior,” adds Enrico Marzotto. These anomalies can be hot (like the plumes rising from the Earth's deep mantle) or cold (like the subducting slabs).

 

Link to Publication: Marzotto, E., Koptev, A., Speziale, S. et al. Olivine’s high radiative conductivity increases slab temperature by up to 200K. Nat Commun 16, 6058 (2025). https://doi.org/10.1038/s41467-025-61148-8

 

 Image: Sketch of subducting slab and transport of water-bearing minerals to the Mantle Transition Zone. Visualisation: Enrico Marzotto.

 

 Journal

Nature Communications

DOI

10.1038/s41467-025-61148-8 

Method of Research

Observational study

Subject of Research

Not applicable

Article Title

Olivine’s high radiative conductivity increases slab temperature by up to 200K

Article Publication Date

7-Jul-2025

 

Researchers found a better way to teach large language models new skills

North Carolina State University

Facebook

XLinkedInWeChatBluesky
Email

Researchers have developed a technique that significantly improves the performance of large language models without increasing the computational power necessary to fine-tune the models. The researchers demonstrated that their technique improves the performance of these models over previous techniques in tasks including commonsense reasoning, arithmetic reasoning, instruction following, code generation, and visual recognition.

Large language models are artificial intelligence systems that are pretrained on huge data sets. After pretraining, these models predict which words should follow each other in order to respond to user queries. However, the nonspecific nature of pretraining means that there is ample room for improvement with these models when the user queries are focused on specific topics, such as when a user requests the model to answer a math question or to write computer code.

“In order to improve a model’s ability to perform more specific tasks, you need to fine-tune the model,” says Tianfu Wu, co-corresponding author of a paper on the work and an associate professor of computer engineering at North Carolina State University. “However, these models are so large that it is not feasible to re-train the entire model. Instead, you want to determine the smallest number of changes necessary to improve the model’s performance. We’ve developed a technique, called WeGeFT (pronounced wee-gift), that represents a significant advance for fine-tuning these large models.”

The big break-through for fine-tuning these large models was called LoRA, which came out in 2022. LoRA works by using mathematical tools to identify a small subset of key parameters that are most likely to improve a model’s performance on a specific task. There have been many attempts to improve upon LoRA, but Wu and his collaborators found these previous efforts either required significantly more computational power to improve performance, or used the same amount of computing power without improving performance.

“WeGeFT builds on LoRA, but incorporates additional mathematical tools that allow us to determine which of the key parameters the model is already familiar with and which parameters the model would need to ‘learn,’” says Wu. “By placing more weight on the truly novel parameters, we are able to improve model performance compared to LoRA without incorporating significant new computational demands.”

In proof-of-concept testing, the researchers found that WeGeFT performed as well as or better than LoRA and its many variants across a variety of downstream tasks: commonsense reasoning, arithmetic reasoning, instruction following, code generation, and visual recognition.

“We think this is a valuable step forward,” Wu says. “We are now exploring ways that WeGeFT could also be used to identify elements of the model that are responsible for harmful outputs, with the goal of improving AI alignment and ‘surgery’ to improve model safety and outputs. We expect that work to be forthcoming.”

The paper, “WeGeFT: Weight-Generative Fine-Tuning for Multi-Faceted Efficient Adaptation of Large Models,” will be presented July 17 at the International Conference on Machine Learning, being held in Vancouver, Canada. Co-corresponding author of the paper is Chinmay Savadikar, a Ph.D. student at NC State. The paper was co-authored by Xi Song, an independent researcher.

This work was done with support from the National Science Foundation under grants 1909644, 2024688 and 2013451; and from the Army Research Office under grants W911NF1810295 and W911NF2210010.

---

Method of Research

Experimental study

Subject of Research

Not applicable

Article Title

WeGeFT: Weight-Generative Fine-Tuning for Multi-Faceted Efficient Adaptation of Large Models

BU creates new language model to help people obtain more accurate answers to science, health questions

PodGPT combines expert knowledge from podcasts with the latest research articles

Boston University School of Medicine

Facebook

XLinkedIn
MessageWhatsAppEmail

(Boston)—The rise of generative artificial intelligence (AI), particularly large language models (LLMs), has marked a transformative shift in data analysis, interpretation and content generation. These models, trained on extensive textual datasets, have demonstrated the ability to generate contextually accurate and linguistically rich outputs, with profound implications for fields such as science and medicine, where models like OpenAI’s GPT-4 have shown remarkable aptitude. However, the full potential of LLMs in science, technology, engineering, mathematics and medicine (STEMM) remains underexplored, particularly in integrating non-traditional data modalities such as audio content.

In a new study, researchers from Boston University introduce a newly created computer program called PodGPT that learns from science and medicine podcasts to become smarter at understanding and answering scientific questions.

“By integrating spoken content, we aim to enhance our model’s understanding of conversational language and extend its application to more specialized contexts within STEMM disciplines,” explains corresponding author Vijaya B. Kolachalama, PhD, FAHA, associate professor of medicine and computer science at Boston University Chobanian & Avedisian School of Medicine. “This is special because it uses real conversations, like expert interviews and talks, instead of just written material, helping it better understand how people actually talk about science in real life.”

Kolachalama and his colleagues collected more than 3,700 hours of publicly available science and medicine podcasts and turned the speech into text using advanced software. They then trained a computer model to learn from this information. Following this, they tested the model on a variety of quizzes in subjects like biology, math, and medicine, including questions in different languages, to see how well it performed. The results demonstrated that incorporating STEMM audio podcast data enhanced their model’s ability to understand and generate precise and comprehensive information.

According to the researchers, this study shows that voice-based content like podcasts can be used to train AI tools. Kolachalama is also a Founding Member of Faculty of Computing & Data Sciences at Boston University, and an affiliate of Hariri Institute of Computing at Boston University.

“This opens the door to using all kinds of audio, like lectures or interviews, to build smarter and more human-like technology. It also shows promise in making science more accessible in many languages, helping people across the world learn and stay informed,” said Kolachalama.

Not only do the researchers believe that this technology will help make scientific and medical knowledge easier to access, but that listening to the conversations of experts in their field will assist people in making more informed decisions about their health and education.

“This could help improve understanding and diagnosis in many health conditions such as Alzheimer's disease, cardiovascular disease, infectious diseases, cancer and mental health. It may also support learning in areas like public health and planetary health,” said Kolachalama.

These findings appear online in the journal npj Biomedical Innovations.

This project was supported by grants from the Karen Toffler Charitable Trust (V.B.K.), the National Institute on Aging’s Artificial Intelligence and Technology Collaboratories (P30-AG073104 and P30-AG073105, V.B.K.), the American Heart Association (20SFRN35460031, V.B.K. & R.A.), Gates Ventures (R.A. & V.B.K.), and the National Institutes of Health (R01-HL159620 [V.B.K.], R01-AG083735 [R.A. & V.B.K.], R01-AG062109 [R.A. & V.B.K.], and U19-AG068753 [R.A.]).

Note to editors: V.B.K. is a co-founder and equity holder of deepPath Inc. and CogniScreen, Inc. He also serves on the scientific advisory board of Altoida Inc. R.A. is a scientific advisor to Signant Health and NovoNordisk. The remaining authors declare no competing interests.

---

Journal

npj Biomedical Innovations

Method of Research

Computational simulation/modeling

Subject of Research

Not applicable

Article Title

PodGPT: an audio-augmented large language model for research and education

Article Publication Date

7-Jul-2025

COI Statement

V.B.K. is a co-founder and equity holder of deepPath Inc. and CogniScreen, Inc. He also serves on the scientific advisory board of Altoida Inc. R.A. is a scientific advisor to Signant Health and NovoNordisk. The remaining authors declare no competing interests.

From position to meaning: how AI learns to read

A study in JSTAT describes the sharp shift in text comprehension strategies during neural network training

Sissa Medialab

LinkedInWeChatBlueskyMessageWhatsAppEmail

The language capabilities of today’s artificial intelligence systems are astonishing. We can now engage in natural conversations with systems like ChatGPT, Gemini, and many others, with a fluency nearly comparable to that of a human being. Yet we still know very little about the internal processes in these networks that lead to such remarkable results.

A new study published in the Journal of Statistical Mechanics: Theory and Experiment (JSTAT) reveals a piece of this mystery. It shows that when small amounts of data are used for training, neural networks initially rely on the position of words in a sentence. However, as the system is exposed to enough data, it transitions to a new strategy based on the meaning of the words. The study finds that this transition occurs abruptly, once a critical data threshold is crossed — much like a phase transition in physical systems. The findings offer valuable insights for understanding the workings of these models.

---

Just like a child learning to read, a neural network starts by understanding sentences based on the positions of words: depending on where words are located in a sentence, the network can infer their relationships (are they subjects, verbs, objects?). However, as the training continues — the network “keeps going to school” — a shift occurs: word meaning becomes the primary source of information.

This, the new study explains, is what happens in a simplified model of self-attention mechanism — a core building block of transformer language models, like the ones we use every day (ChatGPT, Gemini, Claude, etc.). A transformer is a neural network architecture designed to process sequences of data, such as text, and it forms the backbone of many modern language models. Transformers specialize in understanding relationships within a sequence and use the self-attention mechanism to assess the importance of each word relative to the others.

“To assess relationships between words,” explains Hugo Cui, a postdoctoral researcher at Harvard University and first author of the study, “the network can use two strategies, one of which is to exploit the positions of words.” In a language like English, for example, the subject typically precedes the verb, which in turn precedes the object. “Mary eats the apple” is a simple example of this sequence.

“This is the first strategy that spontaneously emerges when the network is trained,” Cui explains. “However, in our study, we observed that if training continues and the network receives enough data, at a certain point — once a threshold is crossed — the strategy abruptly shifts: the network starts relying on meaning instead.”

“When we designed this work, we simply wanted to study which strategies, or mix of strategies, the networks would adopt. But what we found was somewhat surprising: below a certain threshold, the network relied exclusively on position, while above it, only on meaning.”

Cui describes this shift as a phase transition, borrowing a concept from physics. Statistical physics studies systems composed of enormous numbers of particles (like atoms or molecules) by describing their collective behavior statistically. Similarly, neural networks — the foundation of these AI systems — are composed of large numbers of “nodes,” or neurons (named by analogy to the human brain), each connected to many others and performing simple operations. The system’s intelligence emerges from the interaction of these neurons, a phenomenon that can be described with statistical methods.

This is why we can speak of an abrupt change in network behavior as a phase transition, similar to how water, under certain conditions of temperature and pressure, changes from liquid to gas.

“Understanding from a theoretical viewpoint that the strategy shift happens in this manner is important,” Cui emphasizes. “Our networks are simplified compared to the complex models people interact with daily, but they can give us hints to begin to understand the conditions that cause a model to stabilize on one strategy or another. This theoretical knowledge could hopefully be used in the future to make the use of neural networks more efficient, and safer.”

The research by Hugo Cui, Freya Behrens, Florent Krzakala, and Lenka Zdeborová, titled “A Phase Transition between Positional and Semantic Learning in a Solvable Model of Dot-Product Attention”, is published in JSTAT as part of the Machine Learning 2025 special issue and is included in the proceedings of the NeurIPS 2024 conference.

---

Journal

Journal of Statistical Mechanics Theory and Experiment

Method of Research

Data/statistical analysis

Article Title

A Phase Transition between Positional and Semantic Learning in a Solvable Model of Dot-Product Attention

Article Publication Date

7-Jul-2025

 

Mediterranean bacteria may harbor new mosquito solution

American Society for Microbiology

LinkedInWeChatBlueskyMessageWhatsAppEmail

Highlights:

• Mosquitoes that carry pathogens often develop resistance to insecticides.
• Biopesticides offer an ecologically friendly way to control the pests and mitigate resistance, but options are limited.
• Researchers recently identified bacteria in Crete producing metabolites that quickly kill mosquito larvae in lab tests.
• The compounds might be useful for the development of new biopesticides, though developing the right formulations and delivery method remains a challenge.

Washington, D.C.—Mosquito-borne diseases kill more than 700,000 people every year, according to the World Health Organization, and the mosquitos that spread the disease are difficult to control. Most species have developed resistance to all major classes of synthetic insecticides, many of which pose both environmental and health risks.

Biopesticides, derived from living organisms, may mitigate chemical insecticide resistance and offer an environmentally friendly way forward. This week in Applied and Environmental Microbiology, researchers report that bacterial isolates collected from the Mediterranean island of Crete act as insecticides against Culex pipiens molestus mosquitoes, which can transmit human pathogens such as West Nile Virus and Rift Valley Fever Virus. In lab tests, extracts containing metabolites produced by 3 of the isolates killed 100% of mosquito larvae within 24 hours of exposure. 

Those metabolites might guide the development of biopesticides with minimal ecological side effects, the researchers noted. “They degrade more quickly in the environment and therefore don’t accumulate, and they often don’t kill such a wide range of different insect species as chemical insecticides,” said George Dimopoulos, Ph.D., a molecular entomologist and microbiologist at Johns Hopkins University in Baltimore and at the Institute of Molecular Biology and Biotechnology (IMBB) in Crete. He co-led the new study, conducted in Crete, together with molecular biologist John Vontas, Ph.D., at the IMBB.

Dimopoulos’ research focuses on mosquitoes that transmit human pathogens, and over the past 15 years his group has found microbes that produce metabolites that interfere with the pathogens that cause malaria and dengue, and some bacteria that can kill mosquitoes. More recently, they have been investigating mosquito-killing bacteria in the Mediterranean region as part of the MicroBioPest project, funded by the European Union.

For the new work, they collected 186 samples from 65 locations across Crete. The samples included topsoil, soil from around plant roots, plant tissues, water samples and dead insects. They then exposed C. pipiens molestus larvae to water solutions containing some of the most promising isolates found in the samples. More than 100 of the isolates killed all the mosquito larvae within 7 days, and 37 of those killed the larvae within 3 days. Those 37 isolates represented 20 genera, many of which have not previously been identified as potential biopesticides, said Dimopoulos. 

Further analyses showed that the rapid-acting bacteria killed the larvae not through infection but through the production of compounds like proteins and metabolites. This is promising, Dimopoulos noted, because it suggests that an insecticide based on these bacteria would not depend on the microbes staying alive. The findings have implications not only for controlling mosquitoes but also as safe biopesticides to use for controlling agricultural pests. 

The researchers have now begun studying the chemical nature of those insecticidal molecules more closely and identifying whether they are proteins or metabolites. They’re also mapping out the spectrum of pesticidal activity demonstrated by the bacteria, including screening the isolates against other strains of pathogen-bearing mosquitoes and agricultural pest insects. 

Biopesticides often degrade quickly and require multiple applications, Dimopoulos said, and finding the right way to formulate and deliver the compounds will be a challenge in the future. The new study represents the discovery phase. 

“It’s now entering the basic science phase to understand the molecules’ chemical structures and modes of action, and then we’ll shift to a more applied path, really aiming at prototype product development,” he said. “There is a major push toward developing ecologically friendly insecticides.” 
 

###

The American Society for Microbiology is one of the largest professional societies dedicated to the life sciences and is composed of over 37,000 scientists and health practitioners. ASM's mission is to promote and advance the microbial sciences. 

ASM advances the microbial sciences through conferences, publications, certifications, educational opportunities and advocacy efforts. It enhances laboratory capacity around the globe through training and resources. It provides a network for scientists in academia, industry and clinical settings. Additionally, ASM promotes a deeper understanding of the microbial sciences to all audiences.

---

Journal

Applied and Environmental Microbiology