The recent advancement of large language models presents numerous opportunities for teaching and learning. Despite widespread public debate regarding the use of large language models, empirical research on their opportunities and risks in education remains limited. In this work, we demonstrate the qualities and shortcomings of using ChatGPT 3.5 for physics task development by prospective teachers. In a randomized controlled trial, 26 prospective physics teacher students were divided into two groups: the first group used ChatGPT 3.5 to develop text-based physics tasks for four different concepts in the field of kinematics for 10th-grade high school students, while the second group used a classical textbook to create tasks for the same concepts and target group. The results indicate no difference in task correctness, but students using the textbook achieved a higher clarity and more frequently embedded their questions in a meaningful context. Both groups adapted the level of task difficulty easily to the target group but struggled strongly with sufficient task specificity, i.e., relevant information to solve the tasks was missing. Students using ChatGPT for problem posing rated high system usability but experienced difficulties with output quality. These results provide insights into the opportunities and pitfalls of using large language models in education.

Large language models represent a significant advancement in the field of AI. The underlying technology is key to further innovations and, despite critical views and even bans within communities and regions, large language models are here to stay. This position paper presents the potential benefits and challenges of educational applications of large language models, from student and teacher perspectives. We briefly discuss the current state of large language models and their applications. We then highlight how these models can be used to create educational content, improve student engagement and interaction, and personalize learning experiences. With regard to challenges, we argue that large language models in education require teachers and learners to develop sets of competencies and literacies necessary to both understand the technology as well as their limitations and unexpected brittleness of such systems. In addition, a clear strategy within educational systems and a clear pedagogical approach with a strong focus on critical thinking and strategies for fact checking are required to integrate and take full advantage of large language models in learning settings and teaching curricula. Other challenges such as the potential bias in the output, the need for continuous human oversight, and the potential for misuse are not unique to the application of AI in education. But we believe that, if handled sensibly, these challenges can offer insights and opportunities in education scenarios to acquaint students early on with potential societal biases, criticalities, and risks of AI applications. We conclude with recommendations for how to address these challenges and ensure that such models are used in a responsible and ethical manner in education.

Generative AI technologies such as large language models show novel potential to enhance educational research. For example, generative large language models were shown to be capable of solving quantitative reasoning tasks in physics and concept tests such as the Force Concept Inventory (FCI). Given the importance of such concept inventories for physics education research, and the challenges in developing them such as field testing with representative populations, this study seeks to examine to what extent a generative large language model could be utilized to generate a synthetic dataset for the FCI that exhibits content-related variability in responses. We use the recently introduced ChatGPT based on the GPT 4 generative large language model and investigate to what extent ChatGPT could solve the FCI accurately (RQ1) and could be prompted to solve the FCI as if it were a student belonging to a different cohort (RQ2). Furthermore, we study, to what extent ChatGPT could be prompted to solve the FCI as if it were a student having a different force- and mechanics-related preconception (RQ3). In alignment with other research, we found that ChatGPT could accurately solve the FCI. We furthermore found that prompting ChatGPT to respond to the inventory as if it belonged to a different cohort yielded no variance in responses, however, responding as if it had a certain preconception introduced much variance in responses that approximate real human responses on the FCI in some regards.

Recently, the option to use large language models as a middleware connecting various AI tools and other large language models led to the development of so-called large multimodal foundation models, which have the power to process spoken text, music, images and videos. In this overview, we explain a new set of opportunities and challenges that arise from the integration of large multimodal foundation models in education.

Traditionally, strain localization in metallic glasses is related to the thickness of the shear defect, which is confined to the nanometer scale. Using site-specific x-ray photon correlation spectroscopy, we reveal significantly accelerated relaxation dynamics around a shear band in a metallic glass at a length scale that is orders of magnitude larger than the defect itself. The relaxation time in the shear-band vicinity is up to ten times smaller compared to the as-cast matrix, and the relaxation dynamics occurs in a characteristic three-stage aging response that manifests itself in the temperature-dependent shape parameter known from classical stretched exponential relaxation dynamics of disordered materials. We demonstrate that the time-dependent correlation functions describing the aging at different temperatures can be captured and collapsed using simple scaling functions. These insights highlight how a ubiquitous nanoscale strain-localization mechanism in metallic glasses leads to a fundamental change of the relaxation dynamics at the mesoscale.

Using molecular dynamics simulation, we study nanoindentation in large samples of Cu–Zr glass at various temperatures between zero and the glass transition temperature. We find that besides the elastic modulus, the yielding point also strongly (by around 50%) decreases with increasing temperature; this behavior is in qualitative agreement with predictions of the cooperative shear model. Shear-transformation zones (STZs) show up in increasing sizes at low temperatures, leading to shear-band activity. Cluster analysis of the STZs exhibits a power-law behavior in the statistics of STZ sizes. We find strong plastic activity also during the unloading phase; it shows up both in the deactivation of previous plastic zones and the appearance of new zones, leading to the observation of pop-outs. The statistics of STZs occurring during unloading show that they operate in a similar nature as the STZs found during loading. For both cases, loading and unloading, we find the statistics of STZs to be related to directed percolation. Material hardness shows a weak strain-rate dependence, confirming previously reported experimental findings; the number of pop-ins is reduced at slower indentation rate. Analysis of the dependence of our simulation results on the quench rate applied during preparation of the glass shows only a minor effect on the properties of STZs.