AIF-C01 exam prep

Applications of Foundation Models · From chapter 3.1 Designing applications with foundation models

Larkspur has two candidate models for its assistant with similar quality, speed and price in testing. Next year it plans to train the chosen model further on its own question-and-answer examples. Only one of the two supports fine-tuning on Bedrock. Which criterion should decide?

Choose an answer

• A Customization support, because only one model can be fine-tuned
• B Model size, because the larger model always improves with training
• C Prompt caching, because cached prompts replace the need to train
• D Modality, because fine-tuning requires a model that reads images

Fundamentals of GenAI · From chapter 2.1 Generative AI concepts

A model's documentation says it accepts up to 8,000 tokens per request. Larkspur's returns policy is 7,500 words long. A developer says it will fit with room to spare. Why is that claim risky?

Choose an answer

• A The limit counts printed pages, not words, so the policy may be too long
• B Words often split into several tokens, so it may exceed 8,000
• C The limit only applies to the model's reply, not to the input
• D Tokens are counted only for images, so words never hit the limit

Fundamentals of AI and ML · From chapter 1.1 AI concepts and terminology

Larkspur's fraud model is a decision tree trained on past orders. A vendor's photo checker uses a neural network with many layers. Which statement is correct?

Choose an answer

• A Only the photo checker is AI, because it uses a neural network.
• B Both are machine learning, and the photo checker is also deep learning.
• C Deep learning is a separate field, so the photo checker isn't machine learning.
• D Both are generative AI, because both learned from data.