reasoning. As these systems enter practice, they are beginning to reshape neurological workflows in the 10 collection, synthesis, and summarization of clinical information, and may finally offer a realistic path to 11 reducing documentation burden.12 More fundamentally, digitalization in neurology is evolving from the digitization of inputs for neurologic 13 reasoning-clinical notes, neurophysiologic signals, radiologic images-to the use of AI systems that 14 actively participate in that reasoning. Foundation models increasingly function as interpreters, translating 15 the complex, often opaque outputs of specialized predictive algorithms into clinically meaningful 16 neurological insights.The four accepted articles in this second volume reflect this shift from enhanced signal digitalization 18 toward more capable AI models with expanded interpretive roles (Figure 1). Three focus on LLMs 19 or vision-language models-examining how they perform, how they fail, and how prompt design 20 shapes clinically adjacent outputs-while the fourth returns to a foundational theme of digital 21 neurology: remote, continuous measurement as a candidate biomarker in neurodegenerative disease.Together, these contributions illustrate a field moving from isolated digitalization (signal→data) Figure 1. A three-layer framework for AI-enabled neurology, emphasizing the dependencies between data quality, model capability, and interpretation. Progress depends on coordinated advances across all three layers. We suggest that large language models will play a disproportionate role at Layer 3, where complex and high-dimensional model outputs require careful interpretation, contextualization, and clinical validation.Large language models for neurology: A mini review spans all three levels of our framework-better data,
Hier et al. (Wed,) studied this question.