問題文
A chat application must let the user ask follow-up questions such as "and what about last year" and get an answer that accounts for the earlier turns. Which implementation carries that context?
選択肢
- Send an array of role and content objects representing the conversation in chronological order, so past user prompts and assistant responses are part of the request.
- Put the whole conversation into a single system prompt, rebuilt from the client's own copy of the thread before each new question reaches the model.
- Rely on the model to remember the earlier turns, because inference sessions in Snowflake keep per-user state for the duration of the warehouse session and the history is replayed automatically, turn by turn.
- Store the history in a table and let the search service read it, so the retrieval layer supplies the earlier turns as passages alongside the documents it already indexes for the assistant.