AIMentor — AI Expertise Learning Platform

Untimed. Answers are graded when you submit, with explanations and spaced-repetition scheduling.

Question 1easy
Which three ingredients define an agent?
A vector database, embeddings, and a rerankerAn LLM, tools it can call, and a loop that feeds results backA fine-tuned model, a prompt, and a cacheA chain, a retriever, and a generator
Question 2easy
What is the key difference between a chain and an agent?
A chain uses GPT-4 and an agent uses GPT-3.5A chain runs a fixed, predetermined sequence of steps; an agent decides the steps at runtimeA chain can't use toolsThere is no difference
Question 3easy
In the ReAct pattern, what does each turn interleave?
Retrieval and generationReasoning (Thought), acting (Action/tool call), and observing (Observation/result)Training and inferenceEmbedding and indexing
Question 4medium
Why is the loop — not the LLM — what makes an agent an agent?
The loop trains the modelThe loop feeds tool results back so the model can take an action, observe the outcome, and adjust its next stepThe loop reduces token costThe loop replaces the need for tools
Question 5medium
Why is bounding the loop with max_steps (and a cost/time budget) non-negotiable?
It improves answer qualityAn unbounded agent can loop forever, running up latency and a runaway billThe API requires itIt makes tool descriptions unnecessary
Question 6medium
How does an LLM decide which tool to call?
RandomlyFrom the tools' names and descriptions — so vague descriptions cause wrong-tool/wrong-arg errorsBy the alphabetical order of tool namesBy the size of each tool's output
Question 7medium
In the from-scratch ReAct loop, what signals that the agent is finished?
The step counter reaches zeroAn explicit stop marker in the model's reply, e.g. 'Final Answer:'The tool returns an empty stringThe temperature drops to 0
Question 8hard
Which tasks are better served by a CHAIN than an agent? (Select all that apply)
Summarize an email into 3 bulletsClassify a ticket into one of 8 categoriesAnswer a question from docs via retrieve→generateResearch a topic and decide what to cross-check next based on findings
Select all that apply.
Question 9hard
What is the correct default posture when deciding between a chain and an agent?
Always use an agent — it's more capableDefault to a chain; escalate to an agent only when steps depend on runtime resultsAlways use a chain; agents are never worth itFlip a coin
Question 10hard
What are real costs/risks of choosing an agent over a chain? (Select all that apply)
More LLM calls → higher latency and spendNon-determinism that makes it harder to testNew failure modes (loops, wrong tool, bad args)It can no longer use any tools
Select all that apply.
Question 11medium
In the from-scratch agent, what is a 'tool'?
A special model checkpointA plain function (plus a name and description) the model can invoke to act on the worldA type of embeddingA prompt template
Question 12medium
Name the dominant agent pattern (a single word) that interleaves reasoning and tool actions.
Question 13hard
In one word, what should be your DEFAULT choice — chain or agent — when a task's steps are known in advance?

0 / 13 answered

Agent Fundamentals — Practice quiz