Also Worth Noting — 2026-04-03

Also Worth Noting

02 [RAG] 1000+ Medical Imaging Datasets for Foundation Models Project Imaging-X created a comprehensive survey of over 1000 open-access medical imaging datasets. This is crucial because gathering such medical data is challenging due to the need for clinical expertise and strict privacy regulations. This extensive resource will accelerate the development of powerful AI foundation models for medical imaging, potentially leading to better diagnoses and treatments. link

03 [Multimodal] LLMs Predict Alzheimer's with Interpretable Tabular Data Large language models are now used to predict Alzheimer's disease directly from complex tabular biomarker data. This is challenging as clinical biomarker datasets are often small and incomplete, making them difficult for traditional deep learning models. The method provides doctors with an interpretable tool for more accurate and early Alzheimer's diagnosis, even with limited patient data. link

04 [RAG] RAG Quantifies Transplant Guide Discrepancies A framework uses retrieval-augmented language models (RAG) to systematically compare patient education materials across U.S. solid-organ transplant centers. Systematically quantifying substantial variations in complex patient education across many U.S. centers at scale has been a significant challenge, and this framework provides the first scalable and objective method to measure such discrepancies using a five-label consistency taxonomy. By highlighting inconsistencies, this work can help transplant centers improve the clarity and consistency of their patient guidance materials. link

05 [Video Gen] TokenDial: Continuous Video Attribute Control TokenDial is a new framework that allows smooth, slider-style adjustments to specific attributes within existing text-to-video generation models. This system overcomes the challenge of current generators, which struggle to change one attribute (like motion intensity) without also altering the video's identity, background, or overall consistency. This fine-grained control empowers creators to precisely tune elements in generated videos, making it easier to achieve desired visual effects or character actions for various applications. link

06 [Industry] Framework to Align LLM Behavior with Principles Google developed a framework to evaluate if large language models (LLMs) consistently act according to desired behavioral principles. This is challenging because LLMs often say they align with values but don't always do so, requiring deep probing through diverse scenarios. The framework helps developers build more reliable AI that consistently adheres to ethical guidelines and user expectations in real-world use. link