What are parameters and weights in an AI model?

Parameters are the numbers inside an AI model that it learns during training. They store everything the model has "learned" from its data. Weights make up the vast majority of a model's parameters. The other main type is biases, which you will see in the example below.

A simple example

Imagine a very small AI that predicts your exam score from hours spent studying and hours of sleep. It might use a rule like this:

Predicted score = (weight 1 × hours studied) + (weight 2 × hours slept) + bias

• Hours studied and hours slept are the inputs. They change for each student.
• Weight 1, weight 2 and the bias are the parameters. The model learns them during training, then uses the same values for everyone.
• The weights decide how much each input matters. If weight 1 is large, studying has a big effect on the prediction.
• The bias is a starting value that shifts the result up or down. It is also a parameter.

This tiny model has just three parameters. Large AI models like chatbots work on the same idea, but have billions of parameters connected in many layers.

How do the weights get their values?

The model is not programmed with the right numbers. It learns them through training:

1. The model starts with random weights and makes a guess.
2. Its guess is compared with the correct answer.
3. The weights are adjusted slightly to make the error smaller.
4. This repeats millions or billions of times across huge amounts of data.

Over time, the weights settle into values that capture useful patterns, such as which words tend to follow others or what a cat looks like in a photo.

Not to be confused with hyperparameters: these are settings chosen by people before training, such as how many layers the model has or how long training runs. The model does not learn them.

An everyday comparison

Think of a sound mixing desk with billions of tiny knobs. Each knob is a parameter. Training is the process of turning every knob little by little until the output "sounds right". Once training ends, the knob positions are fixed, and that fixed set of numbers is the model.

Why does this matter?

You may hear that one model has "7 billion parameters" and another has "hundreds of billions". More parameters let a model capture more complex patterns, but they also need more data, computing power and energy. Bigger is not automatically better. Smaller, well-trained models can outperform larger ones on many tasks.

Judging what AI tells you

• AI stores patterns, not facts: a model's knowledge is spread across billions of numbers, not kept in a list it can look up. This is why AI can sound confident while being wrong.
• AI learns from its data, including the flaws: if the training data contained mistakes or bias, those patterns become part of the parameters.
• AI knowledge can be out of date: once training ends, the parameters stop changing, so the model may not know about recent events unless it can search the web.
• Check important answers: treat AI output as a starting point and verify it with trusted sources.

Key takeaway: Parameters, mostly weights, are the learned numbers that make up an AI model. They capture patterns from training data, which explains both what AI does well and why its output still needs to be checked.