sentences or answer questions. Generally, the larger the number of parameters, the better the model can capture fine-grained patterns and improve its performance, although this also increases dependence on training data and computational resources. In addition to weights and biases, there are also hyperparameters, which affect how the model learns and makes predictions. Unlike parameters—which are automatically computed during training— hyperparameters are values that model developers or users can manually specify. Weights and Biases Weights and biases are fundamental components of many machinelearning models. When we say that an AI model has billions of parameters, these parameters refer to its weights and biases. Consider a function that represents the relationship between an input x and an output y: y = wx+b. Here, the value w multiplied by x is the weight, and the value b added to the result is the bias. Weights determine how strongly the model considers (more precisely, how strongly it responds to particular patterns or features in) the input data. Biases act as a kind of baseline or offset, pulling the model’s output in a certain direction regardless of the input. The process of training an AI model is essentially the process of adjusting these weights and biases so that they align with the patterns found in the training data. 22 23