AIMentor — AI Expertise Learning Platform

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

Time remaining: 20:00

Question 1easy
What three factors does every model choice trade off?
Color, size, and shapeCapability, cost, and latencyTokens, embeddings, and vectorsTraining, inference, and deployment
Question 2easy
Why should you NOT memorize a model leaderboard?
Leaderboards are secretSpecific model names and prices change constantly; learn the decision framework and check current models at selection timeModels never changeBecause bigger is always better
Question 3medium
A high-volume, simple classification endpoint should favor which model tier?
The frontier/large tier for maximum qualityThe small/fast tier — cost and latency dominate, and the task is simpleA multimodal modelA fine-tuned frontier model
Question 4medium
An overnight autonomous coding agent should favor which tier, and why?
The fast tier, because speed matters mostThe frontier/large tier — capability and correctness dominate and latency barely matters overnightThe cheapest tier, alwaysA text-only small model
Question 5medium
What is model routing (or a cascade)?
Sending every request to the largest modelRouting each task to the cheapest model that can do it well — e.g. a cheap model triages, a larger model handles only hard casesLoad-balancing across identical modelsTraining a router network
Question 6hard
Roughly how much can task-appropriate routing save versus using a large model for everything?
A negligible amountOften an order of magnitude, when the cheap model handles the easy majority and only a minority escalatesIt always costs moreExactly 50% every time
Question 7hard
When should you choose an open-weight model over a closed API model? (Select all that apply)
Data must stay in-house / on your own infrastructureYou need offline or edge deploymentVery high volume where self-hosting is cheaper, or deep fine-tuningYou want top frontier capability with zero ops
Select all that apply.
Question 8medium
What is the main tradeoff of closed/API models?
They can't do reasoningTop capability and zero ops, but per-token cost, vendor dependency, and your data leaves your environment (per the provider's terms)They are always cheaper than open modelsThey require you to own GPUs
Question 9medium
When do you need a multimodal model?
Whenever you want faster responsesWhen the task involves images/documents/charts (e.g. extracting data from an invoice image), not just textOnly for classificationNever — text models handle everything
Question 10hard
For much of the FR/BE enterprise market, why can the deployment platform matter as much as raw model quality?
Azure models are always smarterEnterprise procurement, data residency, and SSO often mandate running models via Azure (or Bedrock/Vertex), making 'can you deploy on the client's tenant?' a hard requirementDirect APIs are banned in the EUDeployment has no effect on model choice
Question 11hard
What should you verify when running a model through a cloud platform (Azure/Bedrock/Vertex) rather than the direct API?
That the model is freeThat the specific model and any needed features are actually available on that platform/region — feature availability can lag the direct APIThat the direct API is disabledNothing — they are always identical
Question 12medium
Name the three-way tradeoff (three words) at the heart of model selection.
Question 13hard
What technique (one word) sends easy tasks to a cheap model and escalates only hard ones to a larger model?

0 / 13 answered

Provider & Model Selection — Final Exam