Search for: All records

ABSTRACT Neuromesodermal progenitors (NMPs) are a vertebrate cell type that contribute descendants to both the spinal cord and the mesoderm. The undifferentiated bipotential NMP state is maintained when both Wnt signaling is active and Sox2 is present. We used transgenic zebrafish reporter lines to live-image both Wnt activity and Sox2 levels in NMPs and observed a unique cellular ratio in NMPs compared to NMP-derived mesoderm or neural tissue. We used this unique signature to identify the previously unknown anatomical position of a progenitor population that gives rise to midline tissues of the floor plate of the spinal cord and the mesodermal notochord. Thus, quantification of the active Wnt signaling to Sox2 ratio can be used to predict and identify cells with neuromesodermal potential. We also developed the auxin-inducible 2 degron system for use in zebrafish to test the temporal role that Sox2 plays during midline formation. We found that ectopic Sox2 in the presence of Wnt activity holds cells in the undifferentiated floor plate/notochord progenitor state, and that degradation of the ectopic Sox2 is required for cells to adopt a notochord fate. 

Medical financial literacy is essential to make smart decisions in healthcare settings and prevent unanticipated financial hardships. Existing literature has shown that young adults often struggle to understand information associated with health insurance and the financial planning necessary for health-related costs. AI-driven chatbots are emerging as educational tools that have the potential to address this issue. This exploratory study examined an AI chatbot aimed at enhancing medical financial literacy among high school students. Participants engaged with the chatbot’s responses to medical financial questions while also rating the clarity, ease of use, trustworthiness, and educational value of the chatbot engagement. Our experiment results supported that the chatbot increased students’ understanding of the financial aspect of healthcare - 76.9 percent of students reported a high degree of understanding, 80.8 percent rated the chatbot’s responses as clear, and 73.1 percent reported they would recommend it to a peer. The responses indicated that students found the chatbot helpful, but suggested that interactive features be added and/or real-world finance features be incorporated into the chatbot. 

Training large language models (LLMs) increasingly relies on geographically distributed accelerators, causing prohibitive communication costs across regions and uneven utilization of heterogeneous hardware. We propose HALoS, a hierarchical asynchronous optimization framework that tackles these issues by introducing local parameter servers (LPSs) within each region and a global parameter server (GPS) that merges updates across regions. This hierarchical design minimizes expensive inter-region communication, reduces straggler effects, and leverages fast intra-region links. We provide a rigorous convergence analysis for HALoS under non-convex objectives, including theoretical guarantees on the role of hierarchical momentum in asynchronous training. Empirically, HALoS attains up to 7.5x faster convergence than synchronous baselines in geo-distributed LLM training and improves upon existing asynchronous methods by up to 2.1x. Crucially, HALoS preserves the model quality of fully synchronous SGD-matching or exceeding accuracy on standard language modeling and downstream benchmarks-while substantially lowering total training time. These results demonstrate that hierarchical, server-side update accumulation and global model merging are powerful tools for scalable, efficient training of new-era LLMs in heterogeneous, geo-distributed environments.