AIF-C01 covers GenAI capabilities and limitations at approximately 20% of the exam. Key topics: hallucination, nondeterminism, context window, knowledge cutoff, and model selection criteria.
---
What GenAI Does Well
Generative AI creates new content — text, images, code, audio. Unlike traditional AI that classifies existing data, generative AI produces novel outputs.
| Capability | Example | |-----------|---------| | Text generation | Email drafts, report summaries | | Code generation | Auto-write Python functions, suggest bug fixes | | Q&A | FAQ responses from documents | | Translation | Real-time multilingual translation | | Image generation | Create images from text descriptions |
---
Hallucination — When AI Makes Things Up
Hallucination is when an AI confidently generates information that is factually incorrect or simply made up.
Think of a friend who doesn't know the answer but confidently gives you wrong information anyway. AI does the same — and unlike the friend, it doesn't even realize it's wrong.
Why does hallucination happen?
Models learn statistical patterns ("in this context, this text typically follows"). They don't memorize facts. When asked about something outside their training data, they generate plausible-sounding content that fits the pattern.
Mitigation methods
| Method | Effect | |--------|--------| | RAG | Most effective — grounds answers in real documents | | Grounding | Connects responses to verified data sources | | Lower Temperature | More conservative, consistent answers | | Human review loop | Person confirms critical decisions |
---
Nondeterminism — Same Question, Different Answers
Unlike a calculator that always gives the same result, an AI can give different answers to the same question. This is nondeterminism.
Temperature controls this
| Temperature | Behavior | Best for | |-------------|----------|---------| | Low (0.0–0.2) | Predictable, consistent | Legal summaries, factual Q&A, code | | Medium (0.3–0.7) | Balanced | General conversation | | High (0.8–1.0) | Creative, varied | Brainstorming, creative writing |
---
Context Window — AI's Working Memory
The context window is the maximum number of tokens the model can process in a single request. If a document exceeds this limit, it must be split into chunks — which is why RAG uses chunking.
---
Knowledge Cutoff
Training data has a cutoff date. The model knows nothing about events after that date. Solution: use RAG to retrieve current information in real time.
---
Model Selection Criteria
| Criteria | Larger Model | Smaller Model | |---------|-------------|--------------| | Accuracy | Higher | Lower | | Speed | Slower | Faster | | Cost | Higher | Lower | | Context window | Generally larger | May be smaller |
---
Exam Key Points
"AI confidently generates false information" — Hallucination "Best solution for hallucination" — RAG "Same question yields different answers" — Nondeterminism "Low Temperature" — Predictable, consistent answers "High Temperature" — Creative, varied answers "Model unaware of events after training date" — Knowledge Cutoff "Maximum tokens in a single request" — Context Window RAG solves both hallucination and knowledge cutoff