AI Fundamentals Overview
Check how a model fits inside a product and what stays true when models and prices change.
3 questions
Take quizCritical thinking for an AI world
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.
Check how a model fits inside a product and what stays true when models and prices change.
3 questions
Take quizTrace what an app really sends with each chat message and which parts the model actually sees and writes.
5 questions
Take quizJudge what a model's weights can and can't know, and where its knowledge and habits come from.
6 questions
Take quizJudge how tokens are counted and billed, and where a growing chat's bill really comes from.
5 questions
Take quizDecide what stays in a model's context window, what leaves, and why fitting isn't the same as using.
5 questions
Take quizJudge what belongs in a system prompt, how much weight it really carries, and what it can't protect.
5 questions
Take quizChoose sampling settings for each task and spot what temperature can't make predictable.
4 questions
Take quizJudge what a JSON schema guarantees about model output and what your own code must still check.
5 questions
Take quizReview how a model asks for tools and what your own code must do with every call it writes.
4 questions
Take quizChoose the smallest model that clears the bar for each job, and spot where small models slip.
5 questions
Take quizDecide when fine-tuning beats a solid prompt, and what limits and side effects a fine-tune brings.
5 questions
Take quizFollow a coding agent's loop of requests and tools, and judge its memory, verification, and safety checks.
5 questions
Take quizJudge how image and video models turn noise or tokens into pictures, and what steers, slows, or breaks them.
6 questions
Take quizCheck how a tensor's shape and data type decide its size, its memory, and the math a model runs on it.
6 questions
Take quizJudge how attention, masking, and position shape what a transformer learns and what long inputs cost.
7 questions
Take quizDiagnose why generated content feels generic or fake, and judge which fixes and detection claims hold up.
6 questions
Take quizWork out why training needs thousands of linked GPUs and how splitting, wiring, and checkpoints keep it running.
6 questions
Take quizDecide when thinking tokens are worth paying for, and judge what visible reasoning can and can't tell you.
6 questions
Take quizSort out what an open model actually shares, what its license allows, and what running it yourself costs.
5 questions
Take quizCheck whether a model will fit and run well on your own machine, and decide when local beats a hosted API.
5 questions
Take quizWeigh what dense and mixture-of-experts designs cost in memory and math, and what model labels really tell you.
6 questions
Take quizDiagnose where made-up answers come from, and choose the fixes and checks that actually make them rarer.
6 questions
Take quizJudge how RAG gets the right passages in front of a model and when a small knowledge base can skip it.
6 questions
Take quizCheck what embedding similarity can and cannot tell you when you search text by meaning.
5 questions
Take quizDecide how a vector database should store, search, and measure embeddings, and what its speed shortcuts give up.
6 questions
Take quizDecide how to cut documents into chunks that keep their context without splitting answers or wasting tokens.
5 questions
Take quizJudge when to pair keyword and vector search and how to merge their rankings without one swamping the other.
4 questions
Take quizDecide which facts belong in hard filters, where their values should come from, and how filters fail quietly.
4 questions
Take quizJudge why a fast first search misorders results and how a careful second pass fits without stalling answers.
4 questions
Take quizDiagnose whether retrieval missed what mattered or dragged in junk, and which of the two you can still fix later.
4 questions
Take quizDecide when to turn what a user typed into a better search query and how to catch rewrites that drift.
4 questions
Take quizJudge where access decisions belong in a RAG system and spot the side doors that still leak private text.
4 questions
Take quizCheck whether cited claims are really supported, using the cheapest test that catches each kind of failure.
4 questions
Take quizDiagnose stale answers and decide how a RAG index should pick up new, edited, and deleted documents.
5 questions
Take quizDecide when a model should drive its own searches and how it should react to what each search returns.
4 questions
Take quizJudge how a shared model gateway should handle keys, per-feature quotas, and routing for every AI feature.
5 questions
Take quizJudge how to send easy requests to a cheaper model without handing hard ones a confident, wrong answer.
5 questions
Take quizDecide when fixed code paths beat letting a model choose its own steps, and what an agent's freedom costs.
4 questions
Take quizCheck how an agent loop should stop when the model never decides it is done, and how to size those limits.
5 questions
Take quizDiagnose what changes when a model's answer arrives in pieces, from buffering hops to errors mid-answer.
4 questions
Take quizDecide when slow AI work belongs in a queue and how to keep workers safe when jobs repeat or keep failing.
5 questions
Take quizDecide which model work can wait for a discounted batch and how to run a big batch without costly mistakes.
4 questions
Take quizJudge which failures should send a request to a backup model and how to keep that backup ready to use.
4 questions
Take quizDecide which agent actions need a person's sign-off and how to build an approval step that actually protects.
5 questions
Take quizJudge what turns a chatbot into an agent, how each step stays grounded, and what changes when nobody is watching.
5 questions
Take quizDecide 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 quizDecide 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 quizChoose between a hand-written agent loop and a framework, and weigh what each one costs as your needs change.
6 questions
Take quizJudge how a harness shapes what a model can do, and pick the right guardrail for each job you hand it.
6 questions
Take quizDecide when splitting work across agents pays off, and how to brief them, share facts, and check their work.
6 questions
Take quizJudge when know-how belongs in a skill and how to build, trigger, and safely install one that checks its own work.
6 questions
Take quizDecide 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 quizJudge tool names, schemas, results, and errors by how a model will read them and where it will guess.
6 questions
Take quizDecide who controls each MCP capability, how servers connect, and how far to trust the servers you install.
5 questions
Take quizDecide when a partner's agent needs A2A, and how to hand off work, resume tasks, and trust its replies.
6 questions
Take quizJudge when a hosted gateway pays off, how to pay for it, and what it quietly changes about requests.
6 questions
Take quizChoose how an agent runs as a service, survives restarts, contains the code it writes, and ships safely.
6 questions
Take quizDiagnose duplicate replies, replayed requests, thread mix-ups, and throttled streaming in a team chat bot.
5 questions
Take quizDecide which stations, checks, and safeguards let a software factory fix issues without eroding trust.
7 questions
Take quizJudge how a small factory filters noisy reports, contains fooled models, and decides what ships unreviewed.
6 questions
Take quizDecide how to split one request's tokens into compartments and how to trim each piece when it overflows.
5 questions
Take quizJudge when a provider can reuse a prompt's saved work and which prompt changes quietly break the match.
5 questions
Take quizJudge when a cache that matches questions by meaning should reuse an answer and when it must call the model.
5 questions
Take quizJudge how per-query and per-user costs really add up and which summary numbers can mislead you.
5 questions
Take quizDiagnose where the wait before an answer's first word goes and which changes actually shorten it.
5 questions
Take quizJudge how sharing each decoding step across requests saves cost and where memory and latency push back.
5 questions
Take quizJudge what rounding a model's weights to fewer bits saves, what it can break, and how to check the cost.
5 questions
Take quizDecide how a system should push back, and what it should drop, when requests arrive faster than it can answer.
5 questions
Take quizJudge how to sample traces, write notes, rank failures, and trace each category to the fix it needs.
5 questions
Take quizDecide what belongs in a golden dataset, how to write its expectations, and when its results stop being trustworthy.
4 questions
Take quizJudge what fixed test sets and live traffic can each reveal about a change, and how to connect the two.
4 questions
Take quizCheck whether an LLM judge deserves trust: how it is scoped, measured against human labels, and kept honest.
5 questions
Take quizDiagnose how prompt and model changes break unrelated behavior, and decide how a test suite should catch it.
4 questions
Take quizJudge whether replies stick to their sources and what a faithfulness score can and cannot tell you.
5 questions
Take quizCheck whether an A/B test of two AI versions is set up, run, and read in a way you can trust.
5 questions
Take quizJudge what thumbs and other user signals really mean, and how to turn them into fixes without gaming them.
4 questions
Take quizDecide how long to wait on a model API, which failures to retry, and when a circuit breaker should stop calling it.
5 questions
Take quizPlan what each AI feature becomes when its model fails, when to step down, and how to keep weaker rungs honest.
5 questions
Take quizDiagnose wrong answers from what the model actually saw, and decide what your traces need to keep.
5 questions
Take quizJudge how to measure, tag, and alert on model spend so that a jump in the bill points to a specific cause.
5 questions
Take quizDecide where personal data may flow, how to replace it so a model can still help, and how to restore it safely.
5 questions
Take quizChoose which checks run around a model, how to build each one, and how to handle the mistakes they make.
6 questions
Take quizJudge how much damage an agent's tools could do, and shrink it with narrow tools, user identity, and limits in code.
5 questions
Take quizJudge why injected instructions work, and design agents where a successful injection can't do much harm.
5 questions
Take quizSpot when an assistant mixes private data, untrusted input, and a way out, and choose where to cut.
5 questions
Take quizDecide what an agent's audit log must record, and how to make it a record that people can trust.
5 questions
Take quiz:: Copyright KIRUPA 2026 //--