Level 1
Explain AI from its origins to widely used tools, including changes across versions.
Who can apply
Open one year after obtaining the Level 2 certificate.
Screen-sharing interview
The examiner checks the computer before asking questions. Respond immediately and complete the requested tasks promptly.
- In Zoom, share your entire screen, not just a window or tab.
- Give the examiner of the shared computer. They first check the ’s display settings to confirm that no additional monitor is connected.
- After the check, answer questions and perform tasks yourself on the shared screen. Using the tools being tested is part of the interview; secretly outsourcing answers to or another person is cheating.
Required programming language
Required at every level, from Level 5 to Level 1.
Shell Scripting
The base language for running the computers that build and run .
This certificate requires at every level, from 5 to 1. Building an AI from scratch — fetching , running , moving results and putting a on a — usually happens on the , and whatever language the model is written in, the runs it, joins the pieces and the work. (Some models can also be built with visual no-code tools, and just using AI in a or app needs no . This certificate asks for the shell because it tests building and automating.)
Learn shell programmingExample design question
How would you build an AI for a specified topic? Explain its purpose, data, tools and how you would verify the result.
This is one design-question example. The interview includes several questions and tasks appropriate to each level, not this question alone.
What to learn
- Study AI origins and the evolution of tools and models.
- Compare model and tool by capabilities and limitations.
- Demonstrate how usage differs by version and .
How you are marked
- Could the candidate connect early AI with later developments?
- Could they distinguish tool changes from model version changes?
- Could they explain and demonstrate version-specific and feature-specific usage?
Exam format
There is no written exam and no practical exam. A single Zoom interview with the entire decides the level. The examiner asks questions suited to the level, you explain right away in your own words, and follow-up questions check how deep your understanding goes.
Practice interview questions
These 20 questions are examples of what the examiner may ask in the interview. Do not try to hit a single correct word — explain your answer out loud. Open each one to see a sample explanation and a follow-up question. They are not official exam questions.
How AI began and its history
-
1
Examiner Explain why the 1956 Dartmouth workshop is called the starting point of history.
Show a sample explanation
Sample explanation At a research workshop held at Dartmouth College in the United States in the summer of 1956 by John McCarthy and others, "Artificial Intelligence" began to be used as the name of a field of study (McCarthy and his co-authors first wrote it in their 1955 proposal). Setting the goal of making machines use language and learn on their own, AI started as a field of research.
Follow-up question Before that, in 1950, what question did Alan Turing ask?
-
2
Examiner Explain what a perceptron is and why research stalled for a while after 1969.
Show a sample explanation
Sample explanation The perceptron, introduced by Frank Rosenblatt in 1958, is a simple artificial neural network that multiplies its inputs by , adds them up and outputs 1 when the sum passes a threshold. In 1969 Minsky and Papert showed in their book Perceptrons that a single-layer perceptron cannot solve problems such as XOR; expectations fell and funding for neural network research shrank.
Follow-up question What later overcame this limit?
-
3
Examiner Explain what an winter is and why it happened more than once.
Show a sample explanation
Sample explanation An AI winter is a period when investment and research funding dropped sharply because high expectations of AI were not met by real results. They came in the mid-1970s and from the late 1980s to the early 1990s. They teach that disappointment follows when promises run ahead of what the technology can actually do.
Follow-up question What should we be careful about so the same thing does not happen in today's AI boom?
-
4
Examiner Explain why the field moved from rule-based to that learns from .
Show a sample explanation
Sample explanation In expert systems people wrote "if ..., then ..." rules by hand; once there were thousands of rules they clashed, were hard to change, and could never cover every exception. As computers and data grew, letting the machine find the rules itself from example data — machine learning — worked better than people writing them.
Follow-up question Where is a rule-based approach still better today?
-
5
Examiner Explain what AlexNet in 2012 and the in 2017 each changed.
Show a sample explanation
Sample explanation AlexNet, a deep trained on , beat other methods by a wide margin in the 2012 ImageNet competition and opened the era. The Transformer, from the 2017 paper " Is All You Need", processes a whole sentence at once with attention and became the basis of today's large .
Follow-up question What role did GPUs play between these two events?
Model generations and versions
-
6
Examiner Explain what changed as the grew from GPT-1 to GPT-3.
Show a sample explanation
Sample explanation GPT-1 (2018) was pre-trained and then fine-tuned for each task; (2019), with 1.5 billion , wrote long, convincing text. GPT-3 (2020) grew to 175 billion parameters and showed few-shot ability: give a few examples in the and it handles a new task.
Follow-up question Can you say that a model keeps getting better just by making it bigger?
-
7
Examiner Explain what made , released in November 2022, different from earlier so that so many people used it.
Show a sample explanation
Sample explanation The base was trained further to follow people's instructions and hold a conversation (instruction tuning and , RLHF). It was also released as a free chat window anyone could try, and estimates said it passed 100 million users within two months.
Follow-up question How does RLHF change a model's answers?
-
8
Examiner Explain what the numbers and dates after a name mean, and why a service pins the version.
Show a sample explanation
Sample explanation Providers keep releasing improved models under the same name. A big number often marks the generation, and a date or small number marks a revision of that generation. An alias that always points to the newest model can change one day and give different results for the same , so a service pins the exact version and switches only after testing the new one.
Follow-up question If the model version you use is scheduled to be retired, what would you prepare?
-
9
Examiner Explain why one generation is released as both large and small , and how to choose between them.
Show a sample explanation
Sample explanation A large model handles hard reasoning better but is slower and more expensive. A small model is fast and cheap, suited to frequent calls such as , summaries and autocomplete. Measure , speed and cost on your real task and choose the smallest model that does well enough.
Follow-up question How would you design a system that sends only the hard questions to the large model?
-
10
Examiner Explain how reasoning (thinking) differ from earlier chat models and when they are worth using.
Show a sample explanation
Sample explanation A reasoning model spends extra thinking through several steps before answering, so it does better on multi-step problems such as maths, and planning. In return it is slower and costs more. Simple questions suit a regular model; hard problems suit a reasoning model or a higher thinking setting.
Follow-up question Does a higher thinking level always give a better answer?
How the tools changed
-
11
Examiner Explain how that helps with programming moved from autocomplete to chat to .
Show a sample explanation
Sample explanation It began as autocomplete suggesting the next line in the editor (for example , released as a preview in 2021). Next, pasting into a chat window and asking became common. Now agents read the project themselves, edit files and run commands and tests. The person's job moved from writing line by line to handing over work and reviewing it.
Follow-up question Which skills have become more important for developers in the age of agents?
-
12
Examiner Explain the difference between using through a web chat and using a (, ).
Show a sample explanation
Sample explanation In a web chat you copy in and carry the answer back yourself, and the AI cannot see your whole project. A terminal reads and edits the files in the folder it runs in and runs commands to check the results. That makes managing and being ready to undo with all the more important.
Follow-up question Are there times when a web chat is the better choice for the same job?
-
13
Examiner Explain how the way connects to outside tools changed, from plugins and to .
Show a sample explanation
Sample explanation At first each service built its own kind of plugin. In 2023 function calling arrived, so could request a tool in a fixed form, and MCP, released by in November 2024, became a shared standard: build a tool once and plug it into many AI apps.
Follow-up question What does a shared standard make better for people who ?
-
14
Examiner Explain why tools have added that keep , such as instruction files, and skills.
Show a sample explanation
Sample explanation A forgets earlier content once a conversation ends, and the it can see at once is limited. So project rules go in instruction files (CLAUDE.md, AGENTS.md), repeated methods go in skills, and user preferences go in memory, so you do not explain them again each time.
Follow-up question What goes wrong when the saved context becomes outdated?
-
15
Examiner Explain what you do when a tool's changes and the old way of using it no longer works.
Show a sample explanation
Sample explanation First check the current version (claude , codex --version), then find the changed commands and settings in the official documentation and change log (, ). Trust the official documentation over old blog posts or an 's , and update your team's documents and to the new usage.
Follow-up question Do you know how to go back to the previous version if an update causes problems?
Usage by version and feature
-
16
Examiner Explain, with a demonstration, when to use a 's plan mode and when to let it act straight away.
Show a sample explanation
Sample explanation Plan mode reads the and shows a plan without changing any files. When the scope is large or the code is new to you, get the plan first, review it, then allow it to proceed; hand over small, clear fixes directly. How to turn it on differs by tool and , so check the help for the version you use.
Follow-up question If the plan you received has mistakes, how would you correct it?
-
17
Examiner Explain how to choose, for a task, between switching the and changing the thinking level.
Show a sample explanation
Sample explanation Many tools let you pick the model with a command such as /model and set the thinking level (reasoning effort) separately. Use a fast, cheap setting for easy fixes, and a large model with a high thinking level for design work or hard . Menu names and options change between , so check them on screen as you demonstrate.
Follow-up question To cut costs while keeping quality, in what order would you try changing the settings?
-
18
Examiner Explain and demonstrate the you use when a conversation gets long (compacting, starting fresh).
Show a sample explanation
Sample explanation As a conversation grows, the fills up, and the misses earlier instructions or slows down. Summarise the conversation to keep only the essentials (/compact in and ), or start a new conversation when the topic changes ( in Claude Code, /new in Codex). Record important decisions in the instruction file so they carry over to new conversations.
Follow-up question What information can you lose by compacting a conversation?
-
19
Examiner Explain what changes when you do the same job with different forms of tool — a , an extension or a desktop app.
Show a sample explanation
Sample explanation A terminal agent is strong at running commands and ; an IDE extension makes it easy to see and choose changes inside the editor; a desktop app is handy for documents and files or for managing several agents. Even tools from the same company can differ in by form, so check the documentation and choose to suit the job.
Follow-up question When team members use different forms of tool, what would you align so the results stay the same?
-
20
Examiner Explain how to run a from a or without a conversation (headless mode), and what to watch out for.
Show a sample explanation
Sample explanation You hand over a job with a single command (for example claude -p "…" or "…"). This lets you repeated work such as scheduled checks or . Since no person approves each step, keep the allowed tools and narrow, and log the results so a person reviews them.
Follow-up question Where and how would you keep the used in CI?