-- online Search!

Critical thinking for an AI world

AI Fundamentals Quizzes

Review how models, retrieval, and agents behave. Catch the assumptions that make an AI feature costly, wrong, or unsafe.

Choose from 86 quizzes below. Practice the concepts you need to judge a solution, recognize faulty assumptions, and find errors in logic or code. Every answer includes an explanation and a link back to the relevant tutorial.

How AI Models Work

AI Fundamentals Overview

Check how a model fits inside a product and what stays true when models and prices change.

3 questions

Take quiz

Anatomy of a Chat Request

Trace what an app really sends with each chat message and which parts the model actually sees and writes.

5 questions

Take quiz

LLM Basics

Judge what a model's weights can and can't know, and where its knowledge and habits come from.

6 questions

Take quiz

Tokens and Billing

Judge how tokens are counted and billed, and where a growing chat's bill really comes from.

5 questions

Take quiz

Context Window Limits

Decide what stays in a model's context window, what leaves, and why fitting isn't the same as using.

5 questions

Take quiz

System vs. User Prompts

Judge what belongs in a system prompt, how much weight it really carries, and what it can't protect.

5 questions

Take quiz

Temperature and Sampling

Choose sampling settings for each task and spot what temperature can't make predictable.

4 questions

Take quiz

Structured Outputs

Judge what a JSON schema guarantees about model output and what your own code must still check.

5 questions

Take quiz

Tool Calling

Review how a model asks for tools and what your own code must do with every call it writes.

4 questions

Take quiz

Small vs. Large Models

Choose the smallest model that clears the bar for each job, and spot where small models slip.

5 questions

Take quiz

Prompting vs. Fine-Tuning

Decide when fine-tuning beats a solid prompt, and what limits and side effects a fine-tune brings.

5 questions

Take quiz

How Coding Agents Work

Follow a coding agent's loop of requests and tools, and judge its memory, verification, and safety checks.

5 questions

Take quiz

The Bigger Picture

Image and Video Generation

Judge how image and video models turn noise or tokens into pictures, and what steers, slows, or breaks them.

6 questions

Take quiz

Tensors, Shapes, and Memory

Check how a tensor's shape and data type decide its size, its memory, and the math a model runs on it.

6 questions

Take quiz

Transformers and Attention

Judge how attention, masking, and position shape what a transformer learns and what long inputs cost.

7 questions

Take quiz

Spotting and Avoiding AI Slop

Diagnose why generated content feels generic or fake, and judge which fixes and detection claims hold up.

6 questions

Take quiz

Training at Data Center Scale

Work out why training needs thousands of linked GPUs and how splitting, wiring, and checkpoints keep it running.

6 questions

Take quiz

AI Reasoning and Thinking Tokens

Decide when thinking tokens are worth paying for, and judge what visible reasoning can and can't tell you.

6 questions

Take quiz

Open Weights and Licenses

Sort out what an open model actually shares, what its license allows, and what running it yourself costs.

5 questions

Take quiz

Running Models Locally

Check whether a model will fit and run well on your own machine, and decide when local beats a hosted API.

5 questions

Take quiz

Dense and Mixture-of-Experts Models

Weigh what dense and mixture-of-experts designs cost in memory and math, and what model labels really tell you.

6 questions

Take quiz

Hallucinations and Made-Up Answers

Diagnose where made-up answers come from, and choose the fixes and checks that actually make them rarer.

6 questions

Take quiz

RAG & Retrieval

RAG Basics

Judge how RAG gets the right passages in front of a model and when a small knowledge base can skip it.

6 questions

Take quiz

Embeddings

Check what embedding similarity can and cannot tell you when you search text by meaning.

5 questions

Take quiz

Vector Databases

Decide how a vector database should store, search, and measure embeddings, and what its speed shortcuts give up.

6 questions

Take quiz

Chunk Size and Overlap

Decide how to cut documents into chunks that keep their context without splitting answers or wasting tokens.

5 questions

Take quiz

Hybrid Search

Judge when to pair keyword and vector search and how to merge their rankings without one swamping the other.

4 questions

Take quiz

Metadata Filtering

Decide which facts belong in hard filters, where their values should come from, and how filters fail quietly.

4 questions

Take quiz

Reranking

Judge why a fast first search misorders results and how a careful second pass fits without stalling answers.

4 questions

Take quiz

Recall vs. Precision

Diagnose whether retrieval missed what mattered or dragged in junk, and which of the two you can still fix later.

4 questions

Take quiz

Query Rewriting

Decide when to turn what a user typed into a better search query and how to catch rewrites that drift.

4 questions

Take quiz

Permission-Aware Retrieval

Judge where access decisions belong in a RAG system and spot the side doors that still leak private text.

4 questions

Take quiz

Citation Grounding

Check whether cited claims are really supported, using the cheapest test that catches each kind of failure.

4 questions

Take quiz

Index Freshness

Diagnose stale answers and decide how a RAG index should pick up new, edited, and deleted documents.

5 questions

Take quiz

Agentic RAG

Decide when a model should drive its own searches and how it should react to what each search returns.

4 questions

Take quiz

AI System Architecture

Model Gateways

Judge how a shared model gateway should handle keys, per-feature quotas, and routing for every AI feature.

5 questions

Take quiz

Routing to Cheaper Models

Judge how to send easy requests to a cheaper model without handing hard ones a confident, wrong answer.

5 questions

Take quiz

Workflows vs. Agents

Decide when fixed code paths beat letting a model choose its own steps, and what an agent's freedom costs.

4 questions

Take quiz

Agent Stop Conditions

Check how an agent loop should stop when the model never decides it is done, and how to size those limits.

5 questions

Take quiz

Streaming Responses

Diagnose what changes when a model's answer arrives in pieces, from buffering hops to errors mid-answer.

4 questions

Take quiz

Queues for Slow AI Jobs

Decide when slow AI work belongs in a queue and how to keep workers safe when jobs repeat or keep failing.

5 questions

Take quiz

Batch vs. Real-Time Inference

Decide which model work can wait for a discounted batch and how to run a big batch without costly mistakes.

4 questions

Take quiz

Fallback Model Chains

Judge which failures should send a request to a backup model and how to keep that backup ready to use.

4 questions

Take quiz

Human Approval Gates

Decide which agent actions need a person's sign-off and how to build an approval step that actually protects.

5 questions

Take quiz

Agents & Integrations

AI Agents

Judge what turns a chatbot into an agent, how each step stays grounded, and what changes when nobody is watching.

5 questions

Take quiz

Context Engineering

Decide what goes into an agent's context window on each trip, and spot windows that hold too little, too much, or a bad fact.

6 questions

Take quiz

Agent Memory

Decide what an agent should remember, where to keep it, when to write and read it, and how to keep it fresh and safe.

6 questions

Take quiz

Frameworks vs. Direct API Calls

Choose between a hand-written agent loop and a framework, and weigh what each one costs as your needs change.

6 questions

Take quiz

Agent Harnesses

Judge how a harness shapes what a model can do, and pick the right guardrail for each job you hand it.

6 questions

Take quiz

Multi-Agent Systems

Decide when splitting work across agents pays off, and how to brief them, share facts, and check their work.

6 questions

Take quiz

Agent Skills

Judge when know-how belongs in a skill and how to build, trigger, and safely install one that checks its own work.

6 questions

Take quiz

Spec-Driven Development

Decide how much spec a job needs, who should review it, and how to turn its criteria into checks that keep agents honest.

6 questions

Take quiz

Designing Tools for Agents

Judge tool names, schemas, results, and errors by how a model will read them and where it will guess.

6 questions

Take quiz

Model Context Protocol

Decide who controls each MCP capability, how servers connect, and how far to trust the servers you install.

5 questions

Take quiz

Agent2Agent Handoffs

Decide when a partner's agent needs A2A, and how to hand off work, resume tasks, and trust its replies.

6 questions

Take quiz

Hosted AI Gateways

Judge when a hosted gateway pays off, how to pay for it, and what it quietly changes about requests.

6 questions

Take quiz

Deploying an Agent

Choose how an agent runs as a service, survives restarts, contains the code it writes, and ships safely.

6 questions

Take quiz

Chatbots Inside Slack

Diagnose duplicate replies, replayed requests, thread mix-ups, and throttled streaming in a team chat bot.

5 questions

Take quiz

AI Software Factories

Decide which stations, checks, and safeguards let a software factory fix issues without eroding trust.

7 questions

Take quiz

Nightowl Case Study

Judge how a small factory filters noisy reports, contains fooled models, and decides what ships unreviewed.

6 questions

Take quiz

Cost & Performance

Token Budgets

Decide how to split one request's tokens into compartments and how to trim each piece when it overflows.

5 questions

Take quiz

Prompt Caching

Judge when a provider can reuse a prompt's saved work and which prompt changes quietly break the match.

5 questions

Take quiz

Semantic Caching

Judge when a cache that matches questions by meaning should reuse an answer and when it must call the model.

5 questions

Take quiz

Cost per Query and User

Judge how per-query and per-user costs really add up and which summary numbers can mislead you.

5 questions

Take quiz

Time to First Token

Diagnose where the wait before an answer's first word goes and which changes actually shorten it.

5 questions

Take quiz

Request Batching

Judge how sharing each decoding step across requests saves cost and where memory and latency push back.

5 questions

Take quiz

Quantization

Judge what rounding a model's weights to fewer bits saves, what it can break, and how to check the cost.

5 questions

Take quiz

Backpressure and Load Shedding

Decide how a system should push back, and what it should drop, when requests arrive faster than it can answer.

5 questions

Take quiz

Evaluation & Quality

Error Analysis on Traces

Judge how to sample traces, write notes, rank failures, and trace each category to the fix it needs.

5 questions

Take quiz

Golden Datasets

Decide what belongs in a golden dataset, how to write its expectations, and when its results stop being trustworthy.

4 questions

Take quiz

Offline vs. Online Evals

Judge what fixed test sets and live traffic can each reveal about a change, and how to connect the two.

4 questions

Take quiz

LLM-as-Judge

Check whether an LLM judge deserves trust: how it is scoped, measured against human labels, and kept honest.

5 questions

Take quiz

Prompt Regression Tests

Diagnose how prompt and model changes break unrelated behavior, and decide how a test suite should catch it.

4 questions

Take quiz

Faithfulness and Groundedness

Judge whether replies stick to their sources and what a faithfulness score can and cannot tell you.

5 questions

Take quiz

A/B Testing

Check whether an A/B test of two AI versions is set up, run, and read in a way you can trust.

5 questions

Take quiz

User Feedback Loops

Judge what thumbs and other user signals really mean, and how to turn them into fixes without gaming them.

4 questions

Take quiz

Reliability & Security

Timeouts, Retries, and Breakers

Decide how long to wait on a model API, which failures to retry, and when a circuit breaker should stop calling it.

5 questions

Take quiz

Graceful Degradation

Plan what each AI feature becomes when its model fails, when to step down, and how to keep weaker rungs honest.

5 questions

Take quiz

Tracing and Prompt Logs

Diagnose wrong answers from what the model actually saw, and decide what your traces need to keep.

5 questions

Take quiz

Token and Cost Metrics

Judge how to measure, tag, and alert on model spend so that a jump in the bill points to a specific cause.

5 questions

Take quiz

PII Redaction

Decide where personal data may flow, how to replace it so a model can still help, and how to restore it safely.

5 questions

Take quiz

Input and Output Guardrails

Choose which checks run around a model, how to build each one, and how to handle the mistakes they make.

6 questions

Take quiz

Least-Privilege Tools

Judge how much damage an agent's tools could do, and shrink it with narrow tools, user identity, and limits in code.

5 questions

Take quiz

Prompt Injection Defense

Judge why injected instructions work, and design agents where a successful injection can't do much harm.

5 questions

Take quiz

The Lethal Trifecta

Spot when an assistant mixes private data, untrusted input, and a way out, and choose where to cut.

5 questions

Take quiz

Audit Logs

Decide what an agent's audit log must record, and how to make it a record that people can trust.

5 questions

Take quiz

The KIRUPA Newsletter

Thought provoking content that lives at the intersection of design 🎨, development 🤖, and business 💰 - delivered weekly to over a bazillion subscribers!

SUBSCRIBE NOW

Creating engaging and entertaining content for designers and developers since 1998.

Follow:

Popular

Loose Ends

:: Copyright KIRUPA 2026 //--