問題文
A colleague asks how a large language model produces a sentence. Which description best explains the core mechanism of a generative language model?
選択肢
- It repeatedly predicts the next token given the tokens produced so far, sampling from a probability distribution over the vocabulary
- It searches an index of documents at inference time and returns the passage with the highest similarity score
- It applies a fixed set of grammar rules to expand a template into a complete sentence
- It compares the input against labeled examples stored in the model and returns the closest matching label