Model
llm, large language model, foundation model, weights, ai modelDefinitions
The trained artifact itself — a set of weights that, given a sequence of tokens, produces a probability distribution over what comes next; the thing that reasons, as distinct from the harness that gives it tools and a loop
The layering is worth holding onto because it assigns blame correctly. A confident wrong answer is the model. A denied permission, a compacted context, a file the agent never saw, a tool that failed silently — harness. Sessions that go badly usually go badly at the second layer, which is the one you can actually change.
In data and domain work, the structure a thing is represented by — a content model, a data model, a domain model
The older sense, and still the more consequential one here. This garden’s content model is its schemas: what a document must declare, what the collection promises, what a page can therefore render. Nuxt Content does not enforce it, so a test does — which makes the model a thing the repository actually holds rather than a diagram.
Two senses that both mean the structure underneath — one learned, one designed.
Avoid: saying "the AI" when you mean one of them. The sentence "the AI deleted my file" is nearly always about a harness that granted a permission, not about a model that wanted to.
Specific models are in the tech registry: the AI models, each with what it reads and writes (modality) and what it is positioned for (model categories).