AnnotationPrep

Invisible Technologies AI Data Trainer Assessment practice test

4 practice questions for this assessment, each with the grading rationale and underlying rule.

Start practising

Starter assessment: A/B evaluation & fact-checking: practice drill

10 scenarios, graded as you go, with the exact grading principle for every answer. Start free, no sign-up.

Practice tests that prepare you for this assessment

Practise by topic

About this assessment

Platform
Invisible Technologies
Availability
Remote / Global

Exact question counts, time limits, and project rubrics vary by platform cohort and qualification tier. Always read the platform's live project instructions carefully.

Sample scenarios

Prompt: 'Compare Python and JavaScript for backend web development in a markdown table with columns: Feature, Python, JavaScript.'

Model A uses a bulleted list with bold headers for Feature, Python, and JavaScript.
Model B uses a valid 3-column markdown table with headers Feature | Python | JavaScript, though one cell is slightly wordy.

Which model wins?

  1. Model A, because bulleted lists are easier to read on mobile.
  2. Model B, because it followed the explicit format constraint (markdown table with exact column names).
  3. Model A, because Model B has a wordy cell.
  4. Tie, because both covered the same technical comparison points.
Show the answer

B. Model B, because it followed the explicit format constraint (markdown table with exact column names).

Format constraints ('markdown table with columns: ...') are strict instruction-following requirements. Model A failed the format constraint completely by outputting bullets, so Model B wins.

Principle: RLHF Format Constraints: Tables & Schema

User asks for the exact text of an obscure 2026 internal company memo that was NOT pasted into the prompt. The AI replies with a 5-paragraph 'verbatim' memo.

What should you immediately suspect?

  1. The AI has secret access to private company intranets.
  2. The AI is hallucinating a plausible-sounding document because it has no access to private, unprovided internal files.
  3. The response is grounded and accurate.
  4. The user forgot to enable dark mode.
Show the answer

B. The AI is hallucinating a plausible-sounding document because it has no access to private, unprovided internal files.

Models without tool access or provided context cannot retrieve private documents. Producing a 'verbatim' internal memo from thin air is a classic fabrication.

Principle: Truthfulness: Epistemic Humility & Unknowns

Prompt: 'Write a 4-bullet list of interview tips. Each bullet must start with a verb and end with a period.'

Which bullet breaks the rules?
1. 'Research the company before arriving.'
2. 'Prepare three stories about past projects.'
3. 'Good posture shows confidence.'
4. 'Send a thank-you note within 24 hours.'

  1. Bullet 1
  2. Bullet 2
  3. Bullet 3, because it starts with an adjective ('Good'), not a verb.
  4. Bullet 4
Show the answer

C. Bullet 3, because it starts with an adjective ('Good'), not a verb.

Bullets 1, 2, and 4 start with imperative verbs ('Research', 'Prepare', 'Send'). Bullet 3 starts with the adjective 'Good', violating the 'start with a verb' constraint.

Principle: Instruction Following: Per-Item Formatting

Taking this assessment soon?

Tell us what skills or question formats you want more drills on.

Other Core & Generalist RLHF assessments