Question 1

How does RAG ensure more accurate responses for specialized queries?

❌ Retraining the model
❌ Increasing token count
❌ Embedding all queries in parameters
✅ Using external knowledge sources to augment the prompt

Explanation:
RAG retrieves relevant domain knowledge and injects it into the prompt, improving accuracy.

---

Question 2

How should Marina ensure an accurate answer about the planet with the most moons?

✅ Use RAG to retrieve the latest information from a reliable source
❌ Embed all moon facts
❌ Rely on pre-trained parameters
❌ Use multiple models for consensus

Explanation:
RAG ensures the answer reflects the most up-to-date scientific data.

---

Question 3

How does retrieval improve LLM accuracy?

❌ Increases model training size
❌ Eliminates external input
❌ Modifies model parameters
✅ Augments the prompt with relevant, up-to-date information

Explanation:
Retrieval pulls relevant chunks from a knowledge base, boosting correctness.

---

Question 4

What should Alex focus on for accurate company policy answers?

❌ Rely only on pre-trained model
❌ Increase token count
✅ Keep the knowledge base up-to-date and well-organized
❌ Embed all possible queries

Explanation:
A clean, current knowledge base ensures accurate retrieval for RAG.

---

Question 5

How does RAG help with a chatbot for complex company mobile policies?

❌ Manual rule-based system
❌ Retrain model
❌ Pre-trained model magically knows
✅ Embed policy documents and retrieve relevant vectors to generate accurate responses

Explanation:
RAG embeds policy docs into vectors and retrieves the most relevant chunks during queries.

---

Question 6

Role of neural networks in embedding:

❌ Identify chunks
❌ Convert text to tokens
✅ Transform token IDs into numeric vectors
❌ Encode tokens into IDs

Explanation:
Embedding models convert token IDs into dense numerical vectors.

---

Question 7

What does tokenization accomplish?

❌ Splits docs into chunks + vectors
❌ Rearranges tokens
❌ Splits vectors into tokens
✅ Splits text into tokens and assigns numerical IDs

Explanation:
Tokenization maps text → tokens → token IDs, which embeddings convert into vectors.

---

🧾 Summary Table