Question 1

How do AI agents differ from chatbots?

Accepted Answer

Chatbots respond to individual prompts in a stateless or lightly stateful conversation. AI agents autonomously plan multi-step tasks, use external tools (APIs, browsers, code execution), and make decisions to achieve a defined goal. An agent can loop, retry, and adapt its approach; a chatbot waits for the next human message.

Question 2

What are LLM hallucinations and how do you reduce them?

Accepted Answer

LLM hallucinations are confident-sounding but factually incorrect outputs generated by a language model. They occur because LLMs predict statistically likely text rather than retrieving verified facts. You reduce them by grounding responses with retrieved source documents (RAG), constraining output with structured schemas, and implementing verification steps.

Question 3

What is inference cost and how do you estimate it?

Accepted Answer

Inference cost is the expense of running a trained model to generate predictions or responses in production. For LLM APIs, it's calculated as (input tokens × input price) + (output tokens × output price) per request. A typical GPT-4-class API call costs $0.01–$0.10 depending on prompt length and response size.

Question 4

What is model context window and why does it matter?

Accepted Answer

A model's context window is the maximum number of tokens (input plus output) it can process in a single request. It determines how much text, code, or conversation history the model can "see" at once. Larger context windows (100K–1M+ tokens) enable processing entire codebases or long documents but increase latency and cost proportionally.

Question 5

What is retrieval-augmented generation (RAG)?

Accepted Answer

Retrieval-augmented generation (RAG) is an AI architecture that combines a large language model with an external knowledge base. Instead of relying solely on training data, the model retrieves relevant documents at query time and uses them to generate more accurate, grounded responses. This reduces hallucinations and keeps answers current without retraining.

Question 6

What is the difference between fine-tuning and prompt engineering?

Accepted Answer

Fine-tuning modifies a model's weights by training it on domain-specific data, permanently changing its behavior. Prompt engineering achieves different outputs by crafting better instructions without altering the model itself. Fine-tuning is costlier and slower but produces more consistent results for specialized tasks; prompt engineering is faster to iterate and requires no training infrastructure.