Fundamentals of Generative AI
Generative AI is the heart of this certification. This domain covers what generative AI is, the foundation models and tokens that power it, key concepts such as embeddings and prompts, and the AWS services, chiefly Amazon Bedrock and Amazon Q, that make generative AI usable. About a quarter of scored questions come from here, so know these concepts cold.
What makes AI generative
Traditional discriminative models classify or score existing inputs, for example labeling an email as spam. Generative AI instead creates new content, text, images, audio, or code, that resembles the data it was trained on. Generative models still require large amounts of training data and significant compute; they are not magic and they do not run for free. Their defining trait is production of novel output rather than selection among fixed categories.
Foundation models and tokens
A foundation model (FM) is a large model pre-trained on broad, massive datasets, giving it general capabilities that can be adapted to many downstream tasks through prompting or fine-tuning, rather than being built for one narrow job. Large language models (LLMs) are foundation models for text. LLMs process text as tokens, chunks such as words or sub-word pieces, and token counts drive both the model's context limit and its pricing. Do not confuse this kind of token with a security token that authorizes an API call.
Embeddings, prompts, and temperature
An embedding is a numeric vector representation of text (or other data) that captures semantic meaning, so items with similar meaning sit close together in vector space; embeddings power semantic search and retrieval. A prompt is the input, instructions, context, and questions, you give a model to steer its output. Temperature is a setting that controls randomness during generation: higher temperature yields more varied, creative output, while lower temperature makes responses more focused and repeatable. A key risk is hallucination, when a model produces fluent, confident text that is actually incorrect or fabricated.
Choosing a foundation model
Selecting a model means balancing capability, output quality, latency, and cost against the needs of the specific task. The largest or newest model is not automatically the best choice; a smaller, cheaper, faster model may fully meet requirements. Understanding these trade-offs, rather than assuming bigger is always better, is exactly the kind of judgment the exam rewards.
Generative AI on AWS
Amazon Bedrock is the managed, serverless service that provides access to foundation models from multiple providers (such as Anthropic, Meta, Cohere, and Amazon) through a single API, so you can build generative applications without managing infrastructure. Amazon Q is AWS's generative AI assistant family: Amazon Q Developer helps write, explain, and debug code and work with AWS, while Amazon Q Business answers questions over enterprise data. Appropriate use cases include drafting content that a human reviews; generative AI should assist, not replace, expert judgment in high-stakes domains.