フリー問題

問題文

An analyst must reduce a column of 90,000 restaurant reviews to a single paragraph describing the most frequent complaints. The combined text is far larger than any model's context window. Which approach works without the analyst writing chunking logic?

選択肢

1. Call AI_AGG on the review column with an instruction describing the aggregation, because this function supports data sets larger than the maximum context window.
2. Call SUMMARIZE on each review and then summarize the summaries by hand, with the second pass seeing only the shortened text and staying inside the window.
3. Sample 500 reviews and summarize only those, treating the sample as representative of the whole month of feedback for the report.
4. Concatenate the reviews with LISTAGG and pass the result to AI_COMPLETE, since a single call avoids the per-row overhead and the model reads the whole text at once, in one pass.