Tag

#LLM

Large language models are the engine under most of what gets called AI today, and these posts try to explain them without either mystifying or dismissing them. I write about how they actually work, the training runs, the tokenization, the RLHF tuning that makes a model polite, and about what they cannot do, which is still a longer list than the marketing admits. The pieces here cover the economics of inference, the open-weight versus closed-weight debate, the context window arms race, and the strange emergent behaviors that appear and disappear between model versions. An LLM is a statistical instrument, powerful and brittle in ways that surprise its own builders. Understanding that brittleness is the prerequisite for using them well.