AWS AIF-C01 Glossary: GenAI, RAG, and Bedrock Terms

AWS AIF-C01 glossary of AI, ML, GenAI, foundation models, and governance terms.

Use this glossary when AI, ML, and generative AI terms start to blur together. Keep it beside the cheat sheet and resources instead of using it as a substitute for study.

High-yield terms

Term Short meaning Why it matters on AIF-C01
Prompt Input instruction given to a model Core generative-AI control surface
Hallucination Fluent but incorrect or unsupported model output One of the most tested generative-AI risks
RAG Retrieval-augmented generation, where a model uses retrieved external context Common “improve grounding without retraining” answer
Fine-tuning Additional training that adapts a model to a narrower task or style Often contrasted with prompting and RAG
Embedding Numeric vector representation of text or other content Core to retrieval and semantic similarity
Vector store System used to store and search embeddings Retrieval layer for many RAG patterns
Guardrail Safety or policy control for model inputs and outputs Strong answer for policy and safety controls
Inference Running a trained or hosted model to produce output Distinct from training or adaptation
Supervised learning Training from labeled examples Foundational ML term the exam still expects
Foundation model Broad pre-trained model adaptable to many tasks Base concept for the Bedrock and GenAI sections
Bedrock AWS managed service for foundation-model access and GenAI building blocks Central AWS service for AIF-C01 service-fit questions
SageMaker AI AWS managed platform for ML development and operations Helps separate foundational ML from GenAI product fit
Token Unit of text a model processes Shows up in cost, context-window, and latency discussions
Context window Amount of input/output context a model can consider in one interaction Helps explain prompt and retrieval limits
Grounding Anchoring model output to trusted source material Strong answer when accuracy or factuality matters
Human-in-the-loop Human review or approval before or after model output is used Common control for high-risk outputs
Prompt injection Malicious or adversarial instruction that tries to override system behavior Important safety concept in GenAI systems
PII Personally identifiable information Core privacy and governance concept

Commonly confused pairs

Pair Keep this distinction clear
prompting vs fine-tuning change the instruction versus change the model behavior
RAG vs fine-tuning retrieve external knowledge versus retrain or adapt the model
model inference vs model training using the model versus creating or updating it
guardrail vs IAM policy model-safety behavior control versus AWS access control
hallucination vs bias unsupported output versus skewed or unfair output pattern
token vs embedding unit of model text processing versus vector representation for similarity and retrieval
Bedrock vs SageMaker AI managed FM access and GenAI building blocks versus broader ML development and operations platform
grounding vs fine-tuning improve answers with trusted retrieved context versus alter model behavior through additional training

If three terms blur together

Cluster Fast separation
prompt / RAG / fine-tuning instruction change, retrieval-backed context, or model adaptation
hallucination / bias / privacy issue wrong answer, unfair pattern, or exposed sensitive data
guardrail / IAM / encryption behavior policy, access control, or data protection
token / embedding / vector store text unit, numeric representation, or storage/search layer
Bedrock / SageMaker AI / prebuilt AI service foundation-model platform, ML development platform, or task-specific managed service

One-sentence memory hooks

  • If the problem is missing trusted knowledge, think RAG before fine-tuning.
  • If the problem is unsafe or policy-violating output, think guardrails plus oversight, not only better prompting.
  • If the question is about business-friendly access to foundation models, think Bedrock.
  • If the issue is broader ML building and operations, think SageMaker AI.
  • If the question is about semantic retrieval, think embeddings plus a vector store.

Operational clusters worth keeping straight

Cluster What it usually signals on the exam
prompting / context / grounding answer quality and factuality controls
RAG / fine-tuning / customization which adaptation strategy matches the requirement
guardrails / human review / monitoring safety and responsible-use controls
IAM / KMS / audit trail access, protection, and governance controls
Bedrock / SageMaker AI / prebuilt AI services service-fit and implementation-scope judgment

If the confusion is really about…

Topic family Best page to revisit
service fit and high-confusion pairs Cheat Sheet
current AWS facts and official prep links Resources
pacing and review order Study Plan
overall exam framing Guide root
RAG, prompting, and FM application choices 3. Applications of Foundation Models
responsible AI and governance controls 4. Guidelines for Responsible AI and 5. Security, Compliance, and Governance for AI Solutions
Revised on Sunday, May 10, 2026