AIF-C01 Domain 2: Generative AI fundamentals

What to understand

A foundation model is trained on broad data and can be adapted to many downstream tasks. Large language models process tokens; embeddings represent meaning as vectors; transformers use attention to model relationships in context.

• Lower temperature generally makes output more predictable; higher temperature increases variation.
• Prompt engineering changes instructions and context without retraining model weights.
• Hallucination is plausible but unsupported output; grounding and verification help reduce its impact.
• Generative AI costs depend on model choice, input/output tokens, latency and throughput.

Common AIF-C01 decisions

Separate model capability from application controls. A model may generate text, images or embeddings, while the application still needs prompts, retrieval, safety controls, monitoring and human review.