Parameters
The internal settings an AI model adjusts as it learns, a bit like the millions of tiny dials on a vast mixing desk. Each one holds a small piece of what the model has picked up from its training, and together they shape every answer it gives. These learned values are called parameters, and their number is often used as a rough measure of a model's size.
Parameters are the values inside an AI model that get tuned during training. You can picture them as an enormous bank of dials. As the model learns from its training data, each dial is nudged a little at a time until, taken all together, they let the model produce sensible text. Nobody sets these values by hand. The model works them out for itself over the course of training.
You will often see model sizes quoted in parameters, usually in billions. A model described as having 70 billion parameters has that many of these adjustable values. As a loose rule, more parameters can mean a more capable model, because there is more room to store what it has learned. It is only a rough guide, though. A smaller, well-trained model can outperform a larger, poorly trained one.
It is worth flagging one point of confusion. In everyday talk about AI, “settings” sometimes get called parameters too, such as the temperature that controls how creative an answer is. Those are different from the internal parameters described here. The learned parameters are baked into the model; the settings are knobs you can turn when you use it.
For a beginner, the main value of the term is reading model comparisons with confidence. When one model is described as larger than another by parameter count, you now know what that number is counting, and that bigger is not automatically better.
Related terms
Used in
- How AI models are trained · Guide