Probability measures how likely an outcome is. In language models, training produces a probability distribution used to predict the next word. section 2.1
Keyboard word suggestions
After selecting “I am,” a keyboard may rank “away” highly and “sandalwood” low. Each suggestion reflects the model’s learned probabilities. section 2.1
These probabilities emerge from unsupervised training on massive datasets, which creates model weights and balances that support prediction. section 2.1
Figure 2.1 Probability distribution in action
Visualizes how candidate words receive different probabilities after preceding text. section 2.1
