AIMentor — AI Expertise Learning Platform

14 questions · pass mark 80% · 20 minute limit. The timer auto-submits at zero.

Time remaining: 20:00

Question 1easy
In a messages/chat API call, what is the `system` role for?
The user's questionsInstructions that set the model's behavior/persona for the whole conversationThe model's previous repliesError messages from the API
Question 2medium
On the Anthropic SDK, what is response.content?
A plain string with the answerA list of content blocks (text, tool_use, …) — filter by block.typeA dictionary keyed by roleThe raw HTTP body
Question 3medium
How do you give a chat model memory of earlier turns?
It remembers automatically across callsResend the full conversation history (including assistant replies) on each callEnable a 'memory: true' flagStore it server-side with a session ID
Question 4medium
Why prefer structured (JSON-schema) outputs over parsing free text?
They are cheaper per tokenYou get valid, typed, parseable data instead of brittle regex parsingThey disable hallucination entirelyThey are required for every API call
Question 5hard
In tool/function calling, who decides WHEN a tool is called and WHAT actually runs?
The model decides when and with what arguments; your code executes the functionYour code decides when; the model executes the functionThe provider runs everything server-side automaticallyThe tool runs itself with no code involved
Question 6medium
What does the SDK's tool runner do?
Runs the model on your GPUAutomates the ask → execute tool → feed result back → continue loopValidates JSON schemas onlyStreams tokens to the browser
Question 7medium
How should API keys be provided to the SDK?
Hard-coded in the source fileFrom an environment variable (e.g. ANTHROPIC_API_KEY), never in codeIn the URL query stringIn the user message
Question 8hard
Best practice for handling API errors?
Catch one broad exception and log the message stringUse typed exceptions, most-specific first, distinguishing retryable (429/5xx) from client errorsString-match the error text to decide what to doIgnore errors; the SDK handles everything
Question 9easy
What does the `assistant` role in the messages array contain?
System instructionsThe model's previous replies, included to give conversational contextThe API keyTool definitions
Question 10medium
Which are true about LLM provider APIs generally? (Select all that apply)
They expose a messages/chat endpoint with role-tagged messagesThey are stateless — history must be resent each callOpenAI, Anthropic, Mistral, and Gemini share the same overall request shapeThey automatically remember every past conversation
Select all that apply.
Question 11medium
messages.parse() returned parsed_output = None. What does that most likely indicate?
The schema was too largeThe model refused (e.g. safety), so no schema-valid output was producedSuccess — None means it workedThe API key is invalid
Question 12easy
When should you stream a response?
Only for classification tasksFor long outputs — to avoid HTTP timeouts and show progressNever; streaming is deprecatedOnly when using tools
Question 13medium
Name the message role (one word) used for instructions that set the model's overall behavior.
Question 14hard
Two words: the API property (as a general concept) meaning the API keeps no memory between calls, so you resend history.

0 / 14 answered

Calling LLM APIs — Final Exam