CBuilderEvals and benchmarks· about 30 min· fundamentals that rarely change· verified 2026-09-20· EN
Evaluate AI performance with your own questions
Be able to construct ten test questions with known answers, assess the AI's responses, and calculate accuracy.
Intuition
How reliable is an AI in your field? Do not guess — measure.
- Formulate ten questions with known answers that you are certain of.
- Ask them, one by one, exactly the same way.
- Assess the answers: correct / incorrect / partial.
- Calculate the proportion of correct answers. Analyse the errors: do they involve numbers, names, or reasoning?
This is a evaluation in miniature — exactly what AI companies do with thousands of questions. Ten questions give a rough picture; a hundred give a better one.
Interactive
Assessment template:
| # | Question | Expected answer | AI's answer | Correct? |
|---|---|---|---|---|
| 1 | What is 15% of 80? | 12 | 12 | ✔ |
| 2 | In which year did World War II end? | 1945 | 1945 | ✔ |
| 3 | Who wrote «Mio, min Mio»? | Astrid Lindgren | Astrid Lindgren | ✔ |
| 4 | How many chromosomes does a human have? | 46 | 48 | ✘ |
Mix simple and complex questions, facts and calculations. Ask the same ten questions to another service — now you have a fair comparison.
Mastery means
- Creates ten questions with verifiable answers and assesses the AI's responses
- Calculates the proportion of correct answers and identifies systematic error patterns
Prerequisites
- CCompare two AI answersrequired
Sign in to do the exercises and build your mastery up.
Sources
- Skolverket — About AI in school (in Swedish) — Skolverket's open terms
- Wikipedia — Benchmark (datavetenskap) (CC BY-SA 4.0) — CC BY-SA 4.0