Question 1

1.1.4 A software testing team needs to evaluate whether a multimodal LLM can handle both textual defect reports and related application screenshots in a single test prompt. What is the key preprocessing step for the image before passing it to the transformer architecture?

Options :

A : Convert it into text via OCR

B : Convert it into embeddings using a vision-language model

C : Directly feed the raw pixel data to the transformer

D : Compress it into a binary format

Answer: B

Question 2

4.1.3 What is the main focus of the hands-on demonstration of an LLM-powered agent?

Options :

A : Measuring energy consumption compared to traditional automation.

B : Demonstrating input data, the agent’s behavior, and its resulting actions.

C : Comparing autonomous vs semi-autonomous performance.

D : Evaluating orchestration as a method to reduce hallucinations.

Answer: B

Question 3

1.1.4 A test engineer needs to identify mismatches between the expected UI layout described in user stories and screenshots from the latest build. They want to maximize test coverage by also generating additional test cases using the same model. Which approach is most suitable?

Options :

A : Use a multimodal LLM with a vision-language module

B : Use a reasoning LLM with chain-of-thought enabled

C : Use separate models for text and images, then combine results manually

D : Use a standard transformer trained only on text data

Answer: A

Question 4

4.2.1 During fine-tuning of an LLM for test generation, the training dataset contains inconsistencies: some user stories are incomplete, and some test cases do not match the described functionality. After fine-tuning, the model frequently generates irrelevant or biased test cases. Which fine-tuning challenge does this illustrate?

Options :

A : Overfitting caused by small dataset size.

B : Opacity in the model’s reasoning.

C : Computational overhead in processing large models.

D : Poor training data quality.

Answer: D

Question 5

2.2.1 A tester provides an LLM with regulatory compliance requirements, historical defect data, and business rules. Which GenAI-supported test analysis task is most directly enabled by this combination?

Options :

A : Identifying ambiguities in GUI layouts

B : Generating non-prioritized test conditions

C : Prioritizing test conditions based on risk level

D : Suggesting test techniques based solely on text patterns

Answer: C