Generative AI Concepts

1 What is Generative AI?

Generative AI refers to AI models that can create new content — text, images, code, audio, video — based on patterns learned from training data. Unlike traditional ML that classifies or predicts, GenAI generates.

Core Concept

Traditional ML: input → prediction (classify, score, predict). Generative AI: prompt → creation (write text, generate image, write code, summarize document). GenAI models are trained on massive datasets and learn to produce new content that follows the patterns of that training data.

2. Key GenAI Concepts

RAG vs Fine-Tuning

RAG = retrieve YOUR documents at query time and feed them to the model as context. Fast, no retraining, data stays up-to-date. Fine-Tuning = retrain the model on YOUR data. Slower, expensive, but deeply customizes model behavior. Exam: "Use company knowledge base without retraining" = RAG. "Customize model for specific domain style" = Fine-Tuning.

3. GenAI on AWS

GenAI Customization Spectrum (least to most effort):

1. PROMPT ENGINEERING (no customization)
   Use foundation model as-is with well-crafted prompts
   Tool: Amazon Bedrock (API calls)

2. RAG (retrieval-augmented generation)
   Combine FM with your knowledge base
   Tool: Amazon Bedrock Knowledge Bases

3. FINE-TUNING (model customization)
   Retrain FM on your data for specialized behavior
   Tool: Amazon Bedrock Custom Models, SageMaker

4. PRE-TRAINING FROM SCRATCH (build your own)
   Train entirely new foundation model
   Tool: Amazon SageMaker (massive GPU clusters)
   Only for: large enterprises, research labs

Exam Tip

GenAI: "Foundation Model" = large pre-trained model (Claude, Titan, Llama). "LLM" = text-specific FM. "RAG" = retrieve docs + generate answer (no retraining). "Fine-tuning" = retrain on your data. "Hallucination" = incorrect but plausible output. "Temperature" = creativity control. "Responsible AI" = fairness, safety, privacy. RAG = fastest way to use your own data with FMs.