.svg)
Kronehoefstraat 85
5612 HL Eindhoven
BTW NL863902509B01
What is it?
A model in AI is a system that has been trained on data to perform a specific task. It stores the learned patterns as millions or billions of parameters and applies them whenever it receives new input.
Models vary considerably in size and specialisation: from small classification models that recognise a type of email, to large language models that write and analyse full texts. In everyday use, 'model' refers both to the technical object and to the product, as in 'GPT-4 is a model from OpenAI'.
Why it matters for SMEs
The model is the core of every AI application. Understanding what a model is helps you set the right expectations for AI tools and assess whether a solution fits your situation.
- Not every model is suited to every task: a model trained on general text performs differently from one specialised in legal documents or accounting data.
- Model choice affects cost and speed: larger models are more capable but more expensive to run and slower to respond. For many SME applications a smaller, faster model is sufficient.
- Models become outdated: the world changes, but a model's training data has a cut-off date. Check what knowledge limit a model has for time-sensitive information.
In practice you rarely choose a model on technical specifications alone. Testing on your own data and tasks gives a more reliable picture of what the model can do in your context.
How it works
A model is built in two phases: training and inference. During training the system learns patterns from large amounts of data. During inference it applies those patterns to new input to produce an outcome.
- Training data is collected and prepared: texts, images, or other inputs with known outcomes.
- The model is trained: it adjusts its internal parameters to minimise the error on the training data.
- The model is validated on new data it has not yet seen.
- After validation the model is deployed: it receives new input and returns predictions or generated output.
As a user or developer you guide how the model applies its learned knowledge through the prompt, the context, and any fine-tuning. The model itself does not change after training unless it is retrained or adjusted.
Example in practice
Picture a construction company wanting to sort incoming quote requests automatically by type of work: renovation, new build, or maintenance. A classification model is trained on historical requests that have been labelled by hand. After training the model processes new requests and automatically assigns a category, so the right colleague or department receives the request without any manual routing.
Comparison and misconceptions
A model is the trained system; an AI application or product is the model plus the surrounding environment, such as the interface, the connection to your data, and the rules that determine when the model is called upon. ChatGPT is a product; GPT-4 is the model beneath it.
.svg){kind=icon}