IBM watsonx Generative AI Engineer - Associate Sample Questions:

1. When addressing bias in a generative AI model, which of the following strategies is least likely to be effective in reducing biased outputs during text generation?

A) Incorporating fairness constraints during the model's training phase
B) Using temperature control during generation to manage diversity in responses
C) Leveraging prompt rephrasing techniques to remove bias-inducing keywords or phrases
D) Training the model on a diverse and representative dataset

2. You want a generative AI model to summarize a lengthy text in one sentence. You provide the following prompt: "Summarize the following paragraph in one sentence: 'Artificial intelligence (AI) is the simulation of human intelligence in machines that are programmed to think and learn. The field of AI includes everything from speech recognition to problem-solving and robotics.'" No prior examples are given.
What type of prompting is being used, and what are the expectations?

A) Zero-shot prompting, but it requires the addition of a few examples for the model to generate a summary.
B) Few-shot prompting, but the model will likely struggle without a few examples of summaries provided.
C) Zero-shot prompting, with the expectation that the model can summarize common concepts like AI due to pre-training.
D) Zero-shot prompting, but the model may fail without detailed instructions on what aspects of the text to focus on.

3. When using IBM Watsonx Tuning Studio, what is the recommended approach to determining the number of training data examples required for effective model fine-tuning?

A) Use at least 50% of the original training data to ensure the fine-tuned model generalizes well across both new and existing tasks.
B) Use a minimum of 1,000 to 5,000 examples for each task, but focus on the quality and relevance of examples rather than quantity.
C) Use at least 10,000 examples for each unique task to ensure the model retains its general knowledge and effectively adapts to the new task.
D) Use no more than 100 examples per task to avoid overwhelming the model's general capabilities with task-specific data.

4. While optimizing the cost of running a Generative AI model, you are instructed to adjust the prompt structure.
Which of the following changes to a prompt would most reduce computational costs while still maintaining effective results?

A) Using stop tokens early in the prompt to minimize generation length.
B) Switching from a narrative-style prompt to a bulleted list format.
C) Including multiple tasks in a single prompt to maximize efficiency.
D) Breaking complex prompts into simpler, sequential prompts.

5. You have successfully deployed a custom model using IBM Watson Machine Learning. Now, an application needs to access the model for inference.
Which of the following configurations will ensure that the application has appropriate access to the deployed model?

A) Create a public-facing API endpoint without authentication
B) Assign the "viewer" role to the application's service account in Watson Machine Learning
C) Implement OAuth2 authentication using IBM Cloud IAM for securing the API.
D) Use the default IAM policies for all users in the IBM Cloud project

Solutions: