Multimodal LLMs and Vision-Language Models CT-GENAI Practice Questions (Page 3)

Questions 11–15

1. 11foundation · easy

   What is a vision-language model (VLM)?

   • AA model that generates only images from text descriptions.
   • BA specific type of multimodal model that integrates visual and textual information.
   • CA model that only processes audio and text data.
   • DA model that translates text from one language to another.
   Reveal Solution Discussion Select an answer first
2. 12application · medium

   A team is using a multimodal LLM to automatically verify that UI elements are correctly aligned in screenshots. The model frequently reports a misalignment that a human tester cannot see. What is the most likely cause of this issue?

   • AThe screenshots are in a format that the model cannot process.
   • BThe model is correctly identifying a defect that is only visible on certain monitors.
   • CThe model's training data did not include any examples of UI screenshots.
   • DThe model is hallucinating a defect that does not exist in the image.
   Reveal Solution Discussion Select an answer first
4. 13expert · hard

   A test manager wants to use a multimodal LLM to automatically verify that a new feature's UI matches the approved design mockups. The team has a strict requirement to avoid any false negatives (i.e., missing a real defect). The LLM has a known tendency to hallucinate. What is the most appropriate strategy to mitigate this risk?

   • AConfigure the LLM to be more 'strict' by setting a lower temperature parameter.
   • BUse the LLM to flag potential discrepancies, but have a human tester review all flagged items before reporting them as defects.
   • CRely solely on the LLM's output, as it is generally accurate.
   • DUse a pixel-diff tool in parallel and only report discrepancies found by both tools.
   Reveal Solution Discussion Select an answer first
5. 14foundation · easy

   Which of the following is a key capability of a multimodal LLM?

   • ACompiling source code into executable binaries.
   • BImage captioning – generating a textual description of an image.
   • CPerforming mathematical calculations with exact precision.
   • DManaging network traffic in a data center.
   Reveal Solution Discussion Select an answer first
6. 15expert · hard

   A QA team is using a multimodal LLM to generate test data for a form with a CAPTCHA. They provide the model with a screenshot of the CAPTCHA and ask it to output the text. The model consistently fails. What is the most likely reason for this failure, and what is the best course of action?

   • AThe screenshot resolution is too low. The team should increase the resolution of the input image.
   • BThe model is hallucinating. The team should re-run the prompt multiple times and take the most common answer.
   • CThe model is not trained to solve CAPTCHAs, which are designed to be difficult for AI. The team should use a dedicated CAPTCHA-solving service or manual input.
   • DThe model is biased against certain types of distorted text. The team should fine-tune the model on a dataset of CAPTCHAs.
   Reveal Solution Discussion Select an answer first