Which function is used to calculate the Manhattan distance between the two points in a vector space?
Answer(s): B
JustificationB: VECTOR_L1_DISTANCE – This function returns the Manhattan (L1) distance between two VARIANT or ARRAY inputs representing points in a multidimensional space. It sums the absolute differences of corresponding coordinates, which is the definition of Manhattan distance. Hence it directly computes the required metric. A. VECTOR_INNER_PRODUCT – Returns the dot product (inner product) of two vectors, not a distance measure. It does not involve absolute differences or summation of magnitudes, so it cannot produce Manhattan distance. C. VECTOR_L2_DISTANCE – Calculates the Euclidean (L2) distance, i.e., the square root of the sum of squared differences. While also a distance function, it corresponds to Euclidean distance, not the Manhattan (L1) distance. D. VECTOR_COSINE_SIMILARITY – Computes the cosine of the angle between two vectors, a similarity metric that relies on dot product and magnitudes, not on absolute coordinate differences. It does not yield a distance value and is unrelated to Manhattan distance.Therefore, B. VECTOR_L1_DISTANCE is the only function that implements the Manhattan distance calculation.
VECTOR_L1_DISTANCE – Snowflake Documentation VECTOR Functions Overview – Snowflake Documentation
A Gen AI Specialist needs to prepare lengthy documents for LLM ingestion. Which Snowflake Cortex function will support this requirement by returning an array of text chunks?
Answer(s): D
JustificationCorrect option – D. SPLIT_TEXT_RECURSIVE_CHARACTER – Returns an array of smaller text chunks by recursively splitting a document at a configurable delimiter. This is exactly the capability needed to prepare lengthy inputs for LLM ingestion, ensuring each chunk fits within token limits while preserving context. Option A. EMBED_TEXT_768 – Generates vector embeddings for a text field; it does not return multiple chunks and therefore cannot be used to fragment a document.Option B. FINETUNE – Triggers model fine-tuning tasks; it does not perform any text partitioning and is unrelated to preprocessing for LLM consumption. Option C. PARSE_DOCUMENT – Extracts semantic structure (headings, tables, etc.) but does not provide a simple array of character-level chunks ; it is oriented toward document understanding rather than chunking for ingestion.Thus, SPLIT_TEXT_RECURSIVE_CHARACTER is the only Snowflake Cortex function that directly satisfies the requirement of returning multiple text chunks ready for LLM processing.
Snowflake Documentation – SPLIT_TEXT_RECURSIVE_CHARACTER : https://docs.snowflake.com/en/sql-reference/functions/split-text-recursive-character Snowflake Documentation – Cortex Functions Overview: https://docs.snowflake.com/en/snowflake-cortex/overviewUse this concise reasoning to reinforce your understanding during exam preparation.
Which commands will restrict access to Snowflake Cortex functions providing access only to roles that have been explicitly-granted privileges? (Choose two.)
Answer(s): C,E
Correct options: C and EC – REVOKE DATABASE ROLE SNOWFLAKE;This statement removes the default DATABASE ROLE named SNOWFLAKE from the Snowflake account. The Cortex engine relies on this database role to expose its functions to callers. By revoking it from the public (implicitly) you eliminate any implicit privilege that would allow unauthenticated roles to invoke Cortex functions. Only roles that are subsequently granted explicit privileges on the underlying database objects can access Cortex.E – REVOKE IMPORTED PRIVILEGES ON DATABASE SNOWFLAKE FROM ROLE PUBLIC;Cortex functions are also exposed through imported privileges that are automatically granted to the PUBLIC role when the database role is created. Explicitly revoking these imported privileges prevents indirect access pathways that bypass the normal role-grant mechanism. After this revocation, a role must receive a direct USAGE or SELECT privilege on the database or schema that hosts the Cortex functions to be able to call them.Why the other options are not appropriateA, B, D – These commands either reference non-existent roles ( SNOWELAK , SNOWFLAK ) or misuse theUSAGE syntax ( REVOKE USAGE ON SCHENA SNOWFLAK; – the object name “SCHENA” is a typo and the statement does not target the Cortex-related database role or schema). They do not affect the privileges that control access to Cortex functions.“CORTEX_FUNCTIONS FROM ROLE PUBLIC;”, “CORTEX_USER FROM ROLE PUBLIC;”, “CORTEX FROM ROLE PUBLIC;” – The Snowflake syntax is REVOKE <privilege> ON <object> FROM ROLE PUBLIC; . The fragments shown are syntactically invalid and would not compile. Moreover, they target generic identifiers ( CORTEX_FUNCTIONS , CORTEX_USER , CORTEX ) that do not map to actual Snowflake securables used byCortex; revoking from PUBLIC on such non-existent objects would have no effect.Repeated “E. CORTEX_USER FROM ROLE PUBLIC;” / “E. CORTEX FROM ROLE PUBLIC;” – Even if the syntax were corrected, revoking a privilege on an object named CORTEX_USER or CORTEX does not correspond to the real securable hierarchy used by Cortex. Only the database-level role and its imported privileges are relevant for controlling access.Conclusion Only C and E correctly remove the default public access path to the Snowflake database role that underpins Cortex and explicitly revoke any imported privileges that could be leveraged by unauthenticated roles. Together they ensure that Cortex functions are usable only by roles that have been expressly granted the required privileges.
1. Snowflake Documentation – Revoke Database Role: https://docs.snowflake.com/en/sql-reference/sql/revoke-database-role 2. Snowflake Documentation – Revoke Imported Privileges: https://docs.snowflake.com/en/sql-reference/sql/revoke-imported-privileges(Both links lead to the official Snowflake SQL command reference pages for REVOKE DATABASE ROLE andREVOKE IMPORTED PRIVILEGES .)
Which Snowflake Cortex feature or function supports the ability of users to ask questions in natural language and receive direct responses without the need to write SQL code?
Why B (Cortex Analyst) is the correct choiceCortex Analyst lets users pose natural-language questions against Snowflake data and receives direct, actionable answers without writing any SQL. It leverages Cortex’s built-in LLM capabilities to translate the user’s intent into an optimized query plan and returns the result in plain language.This capability is purpose-built for interactive, conversational analytics, which matches the question’s description exactly.Why the other options are less suitableA: Cortex Search – Provides full-text or vector search over semi-structured text but does not interpret natural-language questions to generate a response; it only retrieves matching documents. C. CLASSIFY_TEXT – A scalar function that assigns a label to an input text (e.g., sentiment or topic classification); it cannot answer arbitrary questions or retrieve data. D. PARSE_DOCUMENT – Extracts structured information from uploaded documents (e.g., PDFs) but does not support interactive, question-driven querying across the entire Snowflake data lake.References Cortex Analyst – https://docs.snowflake.com/en/snowflake-cortex/cortex-analyst/intro Cortex Search – https://docs.snowflake.com/en/snowflake-cortex/cortex-search/intro
How does Snowflake Cortex Analyst support multi-turn conversations to maintain context across user interactions?
Answer(s): C
Technical justificationThe Cortex Analyst engine is built around stateless LLM calls that receive the complete, token-encoded conversation history (including system prompts, prior user messages, and assistant responses) as part of the input context window. This allows the model to reason over the entire dialogue in a single request, preserving state across multi-turn interactions without needing external storage or session-level tracking on the Snowflake side. Because the full history is supplied at each turn, the analyst can maintain coherence, reference earlier clarifications, and generate responses that are context-aware, which is precisely the capability described in option C.Why the other options are less suitableA: Storing previous interactions in a separate database – Cortex Analyst operates natively within Snowflake and does not require external databases to retain conversation state; doing so would add unnecessary latency and complexity. B. Using a session-based memory feature – Snowflake does not expose a dedicated “session-memory” API for Cortex Analyst; context is managed entirely through the input payload, not via session-scoped attributes. D. Using a summarization agent to rewrite questions – While summarization may be employed for certain downstream tasks, it is not the primary mechanism by which multi-turn context is preserved; the system relies on passing the raw conversation history directly to the LLM agents.
Snowflake Cortex Analyst documentation: https://docs.snowflake.com/en/snowflake-cortical-analyst Multi-turn prompting best practices: https://docs.snowflake.com/en/developer-guide/llm-best-practices#multi-turn-conversations
A: Cortex Search ❌ (Incorrect):Cortex Search is designed to help users search and retrieve relevant data using semantic or keyword-based search capabilities. While it improves data discovery, it does not provide conversational, natural-language question answering with direct responses.B: Cortex Analyst ✅ (Correct):Cortex Analyst allows users to ask questions in natural language and receive direct answers without writing SQL. It interprets user intent, translates questions into underlying queries, and returns results in an easy-to-understand format. This feature is specifically built for natural language interaction with data.C: The Cortex CLASSIFY_TEXT function ❌ (Incorrect):The CLASSIFY_TEXT function is used for text classification tasks, such as categorizing or labeling text based on predefined categories. It does not support natural language question answering over data.D: The Cortex PARSE_DOCUMENT function ❌ (Incorrect):The PARSE_DOCUMENT function is used to extract and structure information from documents. It focuses on document processing, not on answering natural language questions or generating direct responses.
Which strategy should be used to improve latency when using the SNOWFLAKE.CORTEX.COMPLETE function?
Answer(s): A
A: Use the REST API when invoking inferences ✅ (Correct):Using the REST API reduces overhead compared to invoking SNOWFLAKE.CORTEX.COMPLETE through SQL.The REST API avoids query compilation and execution costs associated with SQL, which results in lower end-to-end latency for inference calls. Snowflake specifically recommends the REST API for latency-sensitive Cortex use cases.B: Use the SQL API when using a Snowpark session object ❌ (Incorrect):Using the SQL API through Snowpark does not improve latency. In fact, SQL-based invocation generally adds extra overhead compared to REST-based inference calls.C: Set the temperature parameter close to 1 ❌ (Incorrect):A higher temperature increases randomness and creativity in the output. It does not improve performance and may slightly increase processing time due to more complex generation.D: Pass a JSON schema to the response_format parameter ❌ (Incorrect):Passing a JSON schema helps enforce structured output and simplifies response parsing, but it is not a latency optimization strategy . It improves consistency, not speed.
A Gen AI Specialist needs to use a Snowflake Cortex LLM function to categorize customer questions into two topics: support, and food. The Specialist created a table named QUERIES with these values:Which command should be used?
A ❌ (Incorrect):This command is missing the classification labels. SNOWFLAKE.CORTEX.CLASSIFY_TEXT requires a list of labels (categories) to classify the input text against. Without specifying labels such as support and food, the function cannot perform the classification.B ✅ (Correct):This command uses the correct syntax. SNOWFLAKE.CORTEX.CLASSIFY_TEXT supports positional arguments, where the first argument is the input text and the second argument is the list of labels. This correctly classifies each query into either support or food.C ❌ (Incorrect):The arguments are passed in the wrong order. The function expects the input text first and the labels second. Providing the labels as the first argument causes the function to fail.D ❌ (Incorrect):Although this syntax looks valid, CLASSIFY_TEXT does not require (and does not support in this way for the exam context) explicitly specifying the model using named parameters. The function is intended to be used with the simpler positional-argument format shown in option B.
Post your Comments and Discuss Snowflake GES-C01 exam prep with other Community members:
Our website is free, but we have to fight against AI bots and content theft. We're sorry for the inconvenience caused by these security measures. You can access the rest of the GES-C01 content, but please register or login to continue.