Neural Networks and Deep Learning in Plain Words

A neuron multiplies inputs by weights, adds a bias and applies an activation. Layers stack, so early ones find simple patterns and later ones combine them. Training repeats: guess, measure the loss, adjust the weights. Convolutional networks suit images. Transformers power modern language models.

Weight

A number that says how much an input matters.

Loss

A score of how wrong the model is.

Transformer

The design behind modern language models.