An AI model is essentially a mathematical system trained on data to recognize patterns and make predictions, decisions, or generate outputs. At its core, it’s a function that maps inputs to outputs. What happens in between is learned from data rather than explicitly programmed.
AI models are categorized in multiple overlapping ways: by what they do with data (discriminative vs. generative), by how they learn (supervised, unsupervised, reinforcement, self-supervised), by their internal architecture (transformers, neural networks, diffusion models), and by their scope (narrow vs. foundation models).
Because these dimensions are independent, there’s no single count of model “types”. The taxonomy is fluid and continues to evolve as the field does. For example, a single model like Claude’s Sonnet fits multiple categories at once: generative, self-supervised, transformer-based, and multimodal are all accurate descriptions simultaneously.
Why This Matters for Technical Writers
The model is what your product is built on, and its type shapes what the product can and can’t do. Knowing whether you’re documenting a generative model versus a discriminative one, for instance, affects how you explain outputs, set user expectations, and describe limitations.
It’s also worth understanding that most AI-enabled products today are built on top of foundation models accessed via API. Meaning there’s a layer of product decisions sitting between the model and what users experience.
Some behavior comes from the model itself; some comes from how the product team deployed it. You don’t need to understand the math, but you do need to understand these layers well enough to document behavior accurately and avoid overclaiming.
Common Confusion
“AI model” is often used interchangeably with “AI system,” “AI tool,” or even “AI” itself, but these aren’t the same thing. A model is one component of a larger system that typically includes infrastructure, APIs, safety layers, and a product interface on top.
The model is also frequently confused with the product built on it. ChatGPT is a product; GPT-4 is the model underneath it. Specifically, it’s a multimodal, transformer-based, self-supervised foundation model that functions as both a generative and discriminative system depending on the task.
Related Terms
foundation model, large language model (LLM), generative AI, discriminative model, training data, inference