SnowPro Specialty: Gen AI Question 47
Single answerCortex LLM functions (for example, task-specific, general)A support operations team stores incoming case comments in a Snowflake table. They need a SQL-only solution in Snowflake Cortex to power a dashboard with two outputs for each comment: (1) a concise one-sentence summary for agents, and (2) a sentiment label to help prioritize escalations. The team wants the simplest approach with the least prompt engineering and the most predictable output format for these two specific tasks. Which approach is the BEST fit?
- A
Use task-specific Cortex LLM functions such as SUMMARIZE for the summary and SENTIMENT for the sentiment label directly in SQL.
- B
Use a general-purpose COMPLETE function for both tasks, because general models are always more accurate than task-specific functions.
- C
Export the comments to an external application and call a third-party model endpoint, because Cortex LLM functions do not support summarization or sentiment analysis in SQL.
- D
Use a Snowflake vector search pipeline first, because sentiment classification requires retrieval augmentation before any LLM function can be applied.
Show answer and explanation
Correct answer: A
Explanation
The best answer is to use task-specific Cortex LLM functions for the two defined outputs. In Snowflake Cortex, task-specific functions are intended for common language tasks such as summarization and sentiment analysis, making them a strong fit when teams want SQL-native implementation, low operational overhead, and reduced prompt engineering. A general-purpose completion function is more flexible, but it is typically better suited when custom reasoning, transformation, or bespoke output behavior is required. For narrowly defined tasks, task-specific functions align with Snowflake best practices for simplicity and predictability. This reflects the documented distinction between task-specific Cortex functions and general text-generation/completion capabilities in Snowflake Cortex.
- A. Correct.
Correct. When the requirement is to perform common NLP tasks like summarization and sentiment analysis directly in SQL with minimal prompt engineering, task-specific Cortex LLM functions are the best fit. They are designed for these targeted use cases and typically provide simpler usage and more predictable outputs than prompting a general-purpose model for the same tasks.
- B. Incorrect.
Incorrect. A general-purpose COMPLETE-style function can often perform both tasks, but it usually requires carefully written prompts and output constraints to achieve consistent formatting. The statement that general models are always more accurate is also incorrect. For straightforward tasks such as summarization and sentiment analysis, task-specific functions are usually the simpler and more appropriate choice.
- C. Incorrect.
Incorrect. Snowflake Cortex provides built-in SQL-accessible LLM functions for supported tasks, including summarization and sentiment-related analysis. Exporting data to an external application adds unnecessary operational complexity, latency, governance concerns, and data movement when the use case can be handled natively in Snowflake.
- D. Incorrect.
Incorrect. Vector search and retrieval-augmented generation are useful when a model needs external context from documents or knowledge bases. Sentiment analysis on a single support comment does not inherently require retrieval. Adding a vector search pipeline would not be the simplest or most appropriate solution for this requirement.