Free IBM watsonx Generative AI Engineer v1 - Associate Practice Questions

Sample Practice Questions

Try these 10 sample questions from the IBM watsonx Generative AI Engineer v1 - Associate practice bank. Each includes the answer and explanation. Continue with the interactive practice above to review the full question set.

1. 1A support team wants a model to assign each incoming message to one of several predefined issue categories. Which language task does this require?

   1. A.Text summarization
   2. B.Text classificationCorrect answer: B
   3. C.Text translation
   4. D.Entity extraction

   Explanation: IBM states: "Text classification is a machine learning task that involves assigning predefined labels to text data in order to automatically categorize it into groups." The requested issue categories are predefined labels, so classification addresses the task.

   Reference: What is text classification? — What is text classification?

2. 2A prompt provides several input-and-output demonstrations before asking a model to classify a new message. Which prompting approach is used?

   1. A.Zero-shot prompting
   2. B.Few-shot promptingCorrect answer: B
   3. C.Soft prompt tuning
   4. D.Full model fine-tuning

   Explanation: IBM describes few-shot prompting as "incorporating multiple examples of the task in the prompt." The demonstrations supply those examples at inference time.

   Reference: Simplify Your AI Journey: Unleashing the Power of AI with IBM watsonx.ai — Section 5.1.1, page 59

3. 3Which statement distinguishes a soft prompt used in prompt tuning from a hard prompt?

   1. A.A hard prompt is human-readable text; a soft prompt is learned embeddingsCorrect answer: A
   2. B.A hard prompt is learned embeddings; a soft prompt is human-readable text
   3. C.Both hard and soft prompts are ordinary text written only by a user
   4. D.Both hard and soft prompts are learned embeddings updated by training

   Explanation: IBM describes hard prompts as "prompts written by a human in natural language" and soft prompts as "learnable vector embeddings" concatenated to the hard prompt. The representations differ.

   Reference: What is fine-tuning? — Prompt tuning

4. 4In a retrieval system, what does a text embedding represent?

   1. A.The input text as an array of numerical valuesCorrect answer: A
   2. B.The model's complete trainable weight matrices
   3. C.The tokenizer's integer IDs without a learned representation
   4. D.The training loss recorded at each model epoch

   Explanation: IBM defines vector embeddings as "numerical representations of data points" expressed "as an array of numbers". Text embeddings represent input text numerically for operations such as similarity search.

   Reference: What is vector embedding? — What is vector embedding?

5. 5In IBM's Studio workflow, which comparison correctly describes online and batch deployment?

   1. A.Online serves individual requests; batch processes a collection of inputsCorrect answer: A
   2. B.Online processes only scheduled collections; batch serves immediate requests
   3. C.Both deployment types require all inputs to be scheduled in a batch
   4. D.Both deployment types return only one immediate prediction at a time

   Explanation: IBM says "Online deployment is ideal for real-time processing" and "Batch deployment is suitable for processing large datasets in bulk". Online is appropriate for immediate predictions; batch processes collections.

   Reference: Simplify Your AI Journey: Unleashing the Power of AI with IBM watsonx.ai — Page 43, Online and Batch deployment

6. 6A solution uses Watson Discovery to prepare enterprise documents for downstream AI processing. Which Discovery capability can learn document structure from annotated pages?

   1. A.Smart Document UnderstandingCorrect answer: A
   2. B.A natural-language search query
   3. C.An entity-enrichment result
   4. D.A document retrieval result

   Explanation: IBM says "Smart Document Understanding (SDU) is a visual machine learning tool" that lets users label text so it learns document components such as headers and tables. The other options are query or output artifacts.

   Reference: IBM Watson Discovery — Understands your document structure

7. 7A team wants a shorter overview of a long report that retains its main information. Which language task should the team use?

   1. A.Question answering
   2. B.Text classification
   3. C.Text translation
   4. D.Text summarizationCorrect answer: D

   Explanation: IBM says: "Text summarization condenses one or more texts into shorter summaries for enhanced information extraction." A condensed overview of the report is a summarization task.

   Reference: What is text summarization? — Text summarization defined

8. 8An engineer adds task examples to an inference prompt without running a training job. What happens to the foundation model's parameters as a result of this prompting step?

   1. A.They are updated on each example
   2. B.They are replaced with low-rank adapters
   3. C.They remain unchanged by the promptCorrect answer: C
   4. D.They are quantized for the next response

   Explanation: IBM explains that these prompting techniques use existing model capabilities "without requiring fine-tuning or parameter updates." Supplying examples in the prompt does not itself train the model.

   Reference: Simplify Your AI Journey: Unleashing the Power of AI with IBM watsonx.ai — Section 5.1.1, page 58

9. 9In conventional prompt tuning, which component is trained to adapt the pretrained model to a task?

   1. A.Every weight in the pretrained backbone
   2. B.The learned soft-prompt embeddingsCorrect answer: B
   3. C.The pretrained token embedding table in its entirety
   4. D.Every transformer layer's attention weights

   Explanation: IBM states: "Only the soft prompt parameters and lightweight head (optionally) are updated by using optimizers, while the backbone remains frozen." The soft prompt is adjusted rather than the entire backbone; an optional head may also be trained.

   Reference: What is prompt tuning? — Key components

10. 10When comparing two nonzero embedding vectors, what does cosine similarity primarily measure?

    1. A.The directional alignment of the vectorsCorrect answer: A
    2. B.The straight-line distance between vector endpoints
    3. C.The number of differing positions in binary vectors
    4. D.The overlap between two sets of elements

    Explanation: IBM describes cosine similarity as measuring "how aligned they are in each direction." The same source distinguishes Euclidean distance, Hamming distance and Jaccard similarity.

    Reference: What is a vector database? — Vector search