An AI model is the foundation from which ChatGPT, Gemini and other chatbots are created. The architecture of a model is a mathematical matrix modeled after the human neural system (see
neural network). This is nothing like traditional data processing programming (see
AI programming).
Models are pattern recognition systems that are fine-tuned to answer questions and/or generate original content. However, the model is only part one of a two-part system. The software that actually generates results for people is part two, the "inference engine." It is the inference engine that answers questions, and some inference engines are able to employ multiple models, each trained on a different area; for example, general knowledge or software programming.
Language models deal with text while image models handle drawings and photos. Multimodal models support some combination of text, images, audio and video. See
large language model,
image model and
AI secret sauce.
The AI programmer designs the neural network model with interconnections between each node called "parameters." These parameters start out with preset values and are modified in the training stages to generate better results. Large language models (LLMs) can have billions of parameters and small language models (SLMs) have hundreds of thousands or millions. See
AI training and
AI training vs. inference.