Modern architectures: dense, MoE, long context, multimodal

A · 10-second intuition Most modern LLMs are the same transformer recipe; the differences are where compute is spent per token, how attention scales, and what inputs are encoded.