A team wants a supported text-generation model to produce less varied wording for repeated extraction prompts. What is the most accurate statement about lowering temperature?

1. It generally reduces sampling randomness, but it does not guarantee factual accuracy or perfectly identical results.
2. It increases the model context window so more extraction records fit in every request.
3. It updates the model weights after every prompt so the extraction task becomes permanent.
4. It guarantees that every generated statement is supported by an external source.