Multimodal LLMs and Vision-Language Models CT-GENAI Practice Questions (Page 1)

Questions 1–5

1. 1expert · hard

   A team is building an automated visual regression testing system. They have a pre-trained vision-language model (VLM) that performs well on general image understanding. However, it often fails to detect subtle, application-specific UI changes (e.g., a button's color changing from #FF0000 to #FE0001). The team has a large, labeled dataset of historical UI screenshots and defect reports. What is the most effective strategy to improve the model's performance on this specific task?

   • AUse a larger, more general-purpose multimodal LLM.
   • BFine-tune the VLM on the historical dataset of UI screenshots and defect reports.
   • CIncrease the resolution of the input screenshots.
   • DSwitch to a traditional pixel-diff tool for all visual comparisons.
   Reveal Solution Discussion Select an answer first
2. 2foundation · easy

   Which of the following best describes how a multimodal LLM processes input data?

   • AIt only processes the text portion of any input, ignoring other data types.
   • BIt converts all inputs to a single text format before processing.
   • CIt processes each data type independently and never combines them.
   • DIt aligns and processes different data types together to generate unified responses.
   Reveal Solution Discussion Select an answer first
4. 3application · medium

   A QA team wants to automate the verification of a mobile banking app's login screen across different device resolutions. They have a set of screenshots and a text specification of the expected layout. Which approach best leverages a multimodal LLM for this task?

   • AManually review the screenshots and document any differences in a bug-tracking tool.
   • BFeed both the screenshots and the text specification to a multimodal LLM and ask it to identify visual discrepancies.
   • CUse a computer vision library to extract text from the screenshots and compare it to the specification using a standard text diff tool.
   • DUse a text-only LLM to generate test scripts based on the specification, then run them against the app.
   Reveal Solution Discussion Select an answer first
5. 4application · medium

   A company wants to deploy a multimodal LLM to analyze thousands of UI screenshots nightly. The team is concerned about the cost of running this workload. What is the most significant cost factor they need to consider?

   • AThe cost of labeling the screenshots for training.
   • BThe cost of storing the screenshots.
   • CThe cost of network bandwidth to upload the images.
   • DThe computational resources required to process the images.
   Reveal Solution Discussion Select an answer first
6. 5application · medium

   A test automation engineer is exploring ways to automate the testing of a voice-controlled smart home app. They need a model that can understand the user's spoken command (audio) and the resulting state of the app's interface (image) to verify the correct action was taken. Which type of model is best suited for this end-to-end verification?

   • AA speech-to-text model that transcribes the command and a separate image recognition model.
   • BA vision-language model that processes the image and a text transcript of the audio.
   • CA text-only LLM that processes the log files generated by the app.
   • DA multimodal LLM that can process both audio and image inputs.
   Reveal Solution Discussion Select an answer first