CertifyAIxAce every AI certification.

GES-C02 sample questions with answers

10 free practice questions for the SnowPro Specialty: Gen AI exam. Try each one, then open the answer to see why the right option wins and every other option loses.

Question 1Snowflake for Gen AI Overview

A team embedded 2 million product descriptions last year with the e5-base-v2 model and stored them in a VECTOR column. A new search feature embeds each query with snowflake-arctic-embed-m-v1.5, which returns vectors of the same dimension. Queries run without errors, but the top results are unrelated. What is the most likely cause?

  1. A.

    Query and document vectors come from different models, so their positions in vector space are not comparable

  2. B.

    The stored vectors are more than a year old, and embeddings expire unless the service refreshes them

  3. C.

    The similarity call is placed in the SELECT clause, so it is evaluated after the LIMIT and misses rows

  4. D.

    VECTORCOSINESIMILARITY ignores vector magnitude, so it cannot rank vectors produced by two different models

Show answer

Answer: A

Vectors from different embedding models are not comparable, even when they have the same dimension.

  • A. Correct: similarity is only meaningful between vectors produced by the same embedding model; equal dimensions merely let the function run.
  • B. Incorrect: stored vectors do not expire; they only need regenerating when the text changes or the model changes.
  • C. Incorrect: Snowflake recommends the SELECT clause for the similarity call, with ORDER BY and LIMIT selecting the top matches.
  • D. Incorrect: ignoring magnitude is how cosine similarity works for any vectors; the problem is that the two vector spaces differ.
Question 2Snowflake for Gen AI Overview

A vendor supplies its speech-analytics model only as a container image that runs a proprietary C++ inference server exposing an HTTP API. A retailer wants to run it inside Snowflake next to its data, with Snowflake managing the compute. The Model Registry is proposed. Which route fits the requirement?

  1. A.

    Push the image to a Snowflake image repository and run it as a Snowpark Container Services service

  2. B.

    Log the image to the Model Registry as a built-in model type and call it from a warehouse

  3. C.

    Load the vendor's model weights into Cortex Fine-tuning and call the result with AI_COMPLETE

  4. D.

    Log the image to the Model Registry through CustomModel, then call its methods from SQL

Show answer

Answer: A

A prebuilt, non-Python container image runs in Snowpark Container Services, not through the Python-based Model Registry.

  • A. Correct: SPCS runs any OCI image, in any language or framework, as a managed service on a compute pool, with the image stored in a Snowflake image repository.
  • B. Incorrect: built-in registry types are specific Python frameworks; a C++ server image is not one of them.
  • C. Incorrect: Cortex Fine-tuning adapts supported hosted base models and cannot load a vendor's proprietary model.
  • D. Incorrect: CustomModel wraps Python-serializable models with Python inference code, not a prebuilt non-Python server image.
Question 3Snowflake for Gen AI Overview

A team lead wants Snowflake Copilot inline to follow the team's conventions, such as always using CTEs and fully qualified names, in every suggestion. She searches for a custom instructions setting for Copilot inline in Workspaces and cannot find one. What does Snowflake's documentation say about this?

  1. A.

    Custom instructions for Copilot inline are set by an administrator as an account-level parameter

  2. B.

    Copilot inline reads the custom instructions that each user saved in the Copilot chat panel

  3. C.

    Copilot inline does not accept custom instructions, so conventions must be stated in each request

  4. D.

    Copilot inline takes its custom instructions from the semantic view for the current schema

Show answer

Answer: C

Copilot inline does not accept custom instructions, unlike the Copilot chat panel.

  • A. Incorrect: there is no account parameter that supplies custom instructions to Copilot inline.
  • B. Incorrect: the chat panel's custom instructions apply to that user's chat conversations, and Copilot inline does not accept custom instructions.
  • C. Correct: the Copilot inline documentation states that it does not accept custom instructions to customise how it responds.
  • D. Incorrect: semantic view custom instructions steer Cortex Analyst, not Copilot inline.
Question 4Snowflake for Gen AI Overview

A data scientist sees a Cortex fine-tuned LLM and several registry models on the Snowsight Models page. In a notebook she lists models with the snowflake-ml-python Registry API and uses the same role, but the fine-tuned LLM is not among the models she can manage there. What explains this?

  1. A.

    Her role needs READ on the fine-tuned model before the Registry API will list or return it

  2. B.

    Fine-tuned models become visible to the Registry API only after they are replicated elsewhere

  3. C.

    The Registry API lists only models in the current schema, and the tuned model is in another

  4. D.

    Cortex fine-tuned LLMs appear in the registry's Snowsight UI but are not managed by the registry API

Show answer

Answer: D

Cortex fine-tuned LLMs show on the Snowsight Models page but are managed through Cortex, not the Model Registry API.

  • A. Incorrect: she can already see the model in Snowsight with the same role, so a missing privilege does not explain the difference.
  • B. Incorrect: replication moves a fine-tuned model to another region for inference; it does not change how the registry API treats it.
  • C. Incorrect: the gap is about model type, not schema scope; the model would still not be managed through the API.
  • D. Correct: Snowflake documents that some model types, such as Cortex fine-tuned LLMs, appear in the registry's Snowsight UI but are not managed by the registry API.
Question 5Snowflake for Gen AI Overview

An analyst asks Snowflake Copilot inline to 'return every order whose status means the order was cancelled'. The generated WHERE clause filters on status = 'CANCELLED', but the table actually stores that state as 'X9'. The analyst's role can read the table. Why did Copilot inline guess the wrong value?

  1. A.

    Copilot inline reads column values only from tables that are defined within a semantic view

  2. B.

    Copilot inline only samples the first rows of each table it considers when it writes a query

  3. C.

    Copilot inline cannot read the data in tables, so filter values must be supplied in the request

  4. D.

    The analyst's role lacks the SNOWFLAKE.CORTEX_USER role needed to read values from tables

Show answer

Answer: C

Copilot inline sees names and data types, not table contents, so specific filter values must be given in the request.

  • A. Incorrect: semantic views are a Cortex Analyst concept; Copilot inline does not read values through them.
  • B. Incorrect: Copilot inline does not sample rows at all; it works from object names, data types and the request.
  • C. Correct: Snowflake documents that Copilot has no access to the data inside tables, so you must provide the filter value yourself.
  • D. Incorrect: CORTEX_USER governs access to the feature, and Copilot's lack of data access is by design rather than a missing privilege.
Question 6Snowflake for Gen AI Overview

An engineer creates a personal skill for Cortex Code in her 'etl-dev' workspace to standardise how new tasks are scaffolded, and it works there when she types /. When she opens her 'reporting' workspace and types /, the skill is not listed. What explains this?

  1. A.

    The skill is only listed once her default role has been granted the SNOWFLAKE.CORTEX_USER role

  2. B.

    Personal skills appear in other workspaces only after an administrator approves them for the account

  3. C.

    Personal skills are stored in the workspace where they were created and are only available in that workspace

  4. D.

    Skills created in Snowsight are only available in the CLI until the workspace session is restarted

Show answer

Answer: C

Cortex Code personal skills live in the workspace that created them and are not visible from other workspaces.

  • A. Incorrect: the skill already works in etl-dev under the same user, so role access to Cortex Code is not the issue.
  • B. Incorrect: there is no approval step that publishes a personal skill across workspaces; sharing is done deliberately through skills and plugins.
  • C. Correct: personal skills are stored in the workspace's .snowflake/cortex/skills directory and can only be accessed from that workspace.
  • D. Incorrect: skills created in a Snowsight workspace work in that workspace immediately, as the stem shows.
Question 7Snowflake for Gen AI Overview

Analysts at a hospital network use locked-down laptops on which no software can be installed. They want an AI assistant that knows which SQL file they have open in a workspace, proposes edits they can review in a diff before applying, and can run the resulting queries. Which Cortex Code experience should the platform team enable?

  1. A.

    Snowflake Copilot's chat panel in a legacy SQL worksheet

  2. B.

    Cortex Code in Snowsight, used from Workspaces in the browser

  3. C.

    The Cortex Code CLI connected to VS Code through ACP support

  4. D.

    The Cortex Code CLI, with a shared connections.toml profile

Show answer

Answer: B

With nothing installable locally, the browser-based Cortex Code in Snowsight is the only experience that meets the need.

  • A. Incorrect: the legacy Copilot panel is being replaced by Cortex Code and does not provide diff-reviewed edits to workspace files.
  • B. Correct: Cortex Code in Snowsight is web-based, is aware of the open workspace file, and shows proposed changes in a diff view for review.
  • C. Incorrect: this still requires installing the CLI and an editor locally, which is not permitted.
  • D. Incorrect: the CLI must be installed on the workstation, which the locked-down laptops forbid.
Question 8Snowflake for Gen AI Overview

A team wants a Cortex Search service over support content that must be defined as a join of a tickets table and a knowledge-base table, with some text cleaned up in the source query. A team member starts the Create flow under AI & ML, Cortex Search in Snowsight. Which interface should the team use to define this service, and why?

  1. A.

    The Snowsight flow, run twice so that the two tables each get a service, joined later at query time

  2. B.

    SQL with CREATE CORTEX SEARCH SERVICE, because the Snowsight flow does not handle multiple sources or transformations

  3. C.

    The SEARCH_PREVIEW function, called with a query that joins both tables before they are indexed

  4. D.

    The Snowsight flow, because it accepts a join of two tables when both are chosen on its data source step

Show answer

Answer: B

Define Cortex Search services with SQL when the source needs a join or transformation; the Snowsight flow covers single, untransformed sources.

  • A. Incorrect: two separate services cannot be joined at query time into one search over combined content.
  • B. Correct: Snowflake's instructions say to use SQL if you want to specify multiple data sources or perform transformations when defining the service.
  • C. Incorrect: SEARCH_PREVIEW queries an existing service for testing; it does not define or build a service.
  • D. Incorrect: the Snowsight flow selects a single table, view or stage; it does not build a join of two tables.
Question 9Snowflake for Gen AI Overview

A customer-success team uses a Snowflake Intelligence agent to analyse account health. It now wants users to be able to say 'open a follow-up case for this account' and have the agent create the case through the company's internal logic. Which capability lets the agent take this action?

  1. A.

    A Cortex Analyst verified query that returns the SQL for inserting a new case row

  2. B.

    A second Cortex Search service indexing past cases so that similar ones can be found

  3. C.

    A custom tool on the agent that runs a stored procedure implementing the case logic

  4. D.

    An orchestration instruction telling the agent to write new cases into the case table

Show answer

Answer: C

Agents in Snowflake Intelligence take actions through custom tools that call user-defined functions or stored procedures.

  • A. Incorrect: Cortex Analyst generates read queries for answering questions; it is not a way to perform writes.
  • B. Incorrect: a search service retrieves existing text; it cannot create a case.
  • C. Correct: Snowflake Intelligence's agents support custom tools that execute user-defined functions or stored procedures, which is how an agent takes actions.
  • D. Incorrect: instructions guide the agent's reasoning, but without a tool that performs the action it has no way to create a case.
Question 10Snowflake for Gen AI Overview

An analyst writing long SQL files in Snowsight Workspaces wants the next part of each statement proposed as she types, based on her query history and the tables in the workspace, and accepted with a single key press rather than through a chat request. Which Cortex Code capability provides this?

  1. A.

    AI code suggestions, shown as gray text at the cursor and accepted with Tab

  2. B.

    Quick Edit on highlighted SQL, which rewrites the selection from a prompt

  3. C.

    The Cortex Code chat panel, with the relevant tables added using @ mentions

  4. D.

    The Fix button in the results grid, which proposes a corrected statement

Show answer

Answer: A

AI code suggestions give inline, as-you-type completions in SQL files and notebook SQL cells, accepted with Tab.

  • A. Correct: AI code suggestions predict the next portion of the statement as gray text while typing, using query history and workspace context, and Tab accepts them.
  • B. Incorrect: Quick Edit acts on highlighted text after the analyst asks for a change, not continuously while typing.
  • C. Incorrect: the chat panel is prompt-driven; it does not complete statements inline as the analyst types.
  • D. Incorrect: Fix proposes corrections after a statement has failed; it does not suggest code during typing.

Keep going with 796 more GES-C02 questions

Free papers every day, in the real exam formats, with progress by exam domain. Unlock every paper and timed mock exam when you are ready.

Practise freeSee all GES-C02 papers
GES-C02 sample questions with answers (10 free) · CertifyAIx