Quick answer: The first ten AI Quest levels each teach one core AI-literacy concept: (1) what an LLM is, (2) tokens, (3) training data, (4) hallucination, (5) prompt basics, (6) role-priming, (7) few-shot, (8) output formats, (9) evaluating outputs, (10) human-in-the-loop. Keep notes as you play — each level unlocks a tool you need later.

Level 1: What is an LLM?

First level is identification: match the scenario to “LLM” vs “classical search” vs “rule-based system.” Tip: if the output is creative, open-ended, and language-based, it's almost always an LLM.

Level 2: Tokens

Estimate how many tokens a piece of text will use. Rule of thumb: ~4 characters per token for English. Common gotcha: emojis, code, and non-English scripts use more tokens.

Level 3: Training data

Identify which of three models would know a given fact. The level teaches knowledge cutoffs and retrieval. Common mistake: assuming a model trained in January 2025 knows something from March 2025.

Level 4: Hallucination

Spot the fabricated citation, fake API name, or invented statistic. Tip: when outputs contain numbers or named entities, treat them as claims requiring verification until proven otherwise.

Level 5: Prompt basics

Rewrite a vague prompt into a specific one. Core skill: state the task, the role, constraints, and output format.

Level 6: Role-priming

Assign the right role. A “senior tax accountant” produces different output from a “tax journalist” — same question, different framing. The level rewards matching role to domain.

Level 7: Few-shot

Add two examples to a prompt until the model generalises. Tip: examples should cover the edges of the distribution (not just the easy middle).

Level 8: Output format

Ask for structured output. JSON, CSV, Markdown table. Tip: include an example of the exact format — don't just describe it.

Level 9: Evaluate outputs

Grade three outputs against a rubric. Teaches the difference between “sounds good” and “factually correct.” The AI is confident even when wrong; you are the last line of defence.

Level 10: Human-in-the-loop

Design a workflow where AI does the grunt work and a human approves critical steps. This is the single most important pattern for high-stakes AI use — and the one the EU AI Act formally requires for high-risk systems.

Practice plan

10 minutes a day for 10 days will get you through levels 1–10 and into the intermediate tier. Keep a notebook of the prompts you wrote for later reuse. GeraJobs listings increasingly expect applicants to demonstrate AI-literacy scores.

How to Beat Your First 10 AI Quest Levels (2026 Walkthrough)