Artificial Intelligence & AI Safety
Not just as a tool, but as a field with deep unresolved questions about alignment, capability, and what it means for a system to 'understand' something. The gap between benchmark performance and genuine reasoning is where I think the interesting work lives.