Course Test

Question 1

What is the fundamental difference between LLM training and LLM inference?

Select one answer

Training uses GPU clusters; inference uses CPU clusters

Training adjusts model weights using a corpus; inference runs the frozen model on new input

Training is fast and cheap; inference is slow and expensive

Training happens on your prompt; inference happens on the training dataset

Question 2

An LLM generates a Meshery design manifest that includes a component field named `spec.meshSync.autoDiscovery.enabled`. You cannot find this field in the Meshery documentation. What is the most likely explanation?

Select one answer

The field exists but is undocumented

The field was removed in a recent Meshery release

The LLM hallucinated a plausible-looking field that does not actually exist

The field exists only when MeshSync is installed separately

Question 3

Which of the following approaches correctly addresses the problem of an LLM having no knowledge of your live cluster state?

Select all that apply

Asking the LLM to predict cluster state based on your description of the system design

Passing the output of mesheryctl system check into the prompt before asking questions about it

Using a tool-equipped agent that calls Meshery APIs to retrieve live state at query time

Using a model with a larger context window, which gives it access to more recent information

Question 4

You are designing an automated pipeline where an LLM generates a Meshery design, and the design is applied to your production cluster immediately without human review. Which failure mode makes this architecture most risky?

Select one answer

The LLM will refuse to generate YAML for production environments

The LLM may generate invalid or dangerous configurations with the same confident tone as correct ones

The LLM cannot generate valid YAML without access to the Kubernetes API

The context window will always be exceeded by production-scale manifests

Question 5

What does it mean that LLM output is non-deterministic, and what is the correct mitigation for infrastructure automation?

Type your answer below

Question 6

You need an LLM to analyze the output of mesheryctl system check and identify the root cause across several warnings, then recommend which configuration change to make. Which model tier is most appropriate and why?

Select one answer

A small, fast model - because low latency is more important than accuracy for incident response

A mid-tier model for classification, because the task is just categorizing warnings

A larger, higher-capability model - because multi-step root cause reasoning is where capability gaps are most pronounced

Model tier does not matter for this task because mesheryctl output is always structured