A model with one trillion learnable values that the neural network adjusts during training to improve performance on language tasks.
World knowledge accuracy, recall of facts and relationships
Multi-step reasoning, logic puzzles, mathematical problem-solving