Fundamentals of Generative AI – AIF-C01 Study Guide
This domain accounts for 24% of the exam and covers what generative AI is, its core vocabulary, its capabilities and limitations, and the AWS services for building generative AI. The exam rewards precise vocabulary (tokens, embeddings, foundation models) and knowing which AWS service does what (Bedrock, Amazon Q, SageMaker JumpStart). No implementation depth is required.
Generative AI vs Traditional ML
Traditional (discriminative) ML predicts a label or value from input. Generative AI creates new content — text, images, code, audio — built on Transformer-based foundation models trained on vast data. The exam cue: is the system creating something new (a draft, an image) versus classifying/predicting? Creation → generative AI.
Foundation Models and Core Vocabulary
A foundation model (FM) is a large model pre-trained on broad data that serves as a general-purpose base, adaptable to many tasks via prompting, RAG, or fine-tuning.
| Term | Meaning |
|---|---|
| Token | A word/sub-word unit the model processes; billing and context limits are measured in tokens |
| Prompt | The input instructions/context you provide |
| Completion | The model's generated output |
| Embedding | A numeric vector capturing semantic meaning; enables similarity search (basis of RAG) |
| Context window | Maximum tokens (prompt + completion) a model considers at once |
Why tokens matter: cost is typically per token, and each model has a token-based context limit — so token counts drive both spend and how much text fits.
Unlock with a paid plan
4 more sections (about 300 words) continue below. Every paid plan unlocks all AIF-C01 study guides, the full question bank and mock exams, from $4.97/month.