Learning Path Exam

Question 1

At its core, what does a large language model do?

Select one answer

Looks answers up in a database

Predicts the next token given the preceding context

Executes code deterministically

Compiles YAML into binaries

Question 2

What is the 'context window'?

Select one answer

The model's training dataset

The maximum amount of text (tokens) the model can consider at once

A UI panel in the terminal

The GPU memory size

Question 3

Which are real LLM failure modes to guard against in operations? (Select all that apply.)

Select all that apply

Hallucinating commands or fields that do not exist

Being confidently wrong

Relying on stale training knowledge

Producing perfectly deterministic output every time

Question 4

Why ground a prompt in live cluster/Meshery state instead of relying on the model's memory?

Select one answer

It makes the response longer

Current, specific state yields accurate, relevant answers; training data is stale

It disables hallucination entirely

It is required by Kubernetes

Question 5

Which instruction most reliably yields a parseable design artifact from an LLM?

Select one answer

'Explain your reasoning in detail'

'Output only valid YAML, no commentary'

'Be creative with the format'

'Use as many tokens as possible'

Question 6

Describe the agentic loop in three words (one common phrasing).

Type your answer below

Question 7

What distinguishes a coding agent from a plain chat assistant?

Select one answer

It can take actions via tools (files, shell, APIs), not just produce text

It uses a larger font

It never makes mistakes

It runs without any model

Question 8

Why keep a human in the loop for agent-driven infrastructure changes?

Select one answer

To slow the agent down for no reason

To review and approve changes before they apply, bounding blast radius

Because agents cannot run commands

To increase token usage

Question 9

In retrieval-augmented generation for ops, what is 'retrieval'?

Select one answer

Fetching relevant current information and supplying it to the model as context

Retraining the model on your data nightly

Deleting old logs

Compressing container images

Question 10

Before trusting an infrastructure agent with real changes, you should evaluate it. Which practices help? (Select all that apply.)

Select all that apply

Define golden tasks with known-good outcomes

Run an eval harness across fixtures

Gate real changes behind dry-runs and approvals

Skip all testing to move faster