HOW LLMS ACTUALLY WORK

Summary from the TLDR AI newsletter

Modern large language models are mostly built by stacking transformer blocks over and over - the differences come from what each one was trained on, the scale and configuration choices, and the post-training done on top.