Training Turk

Propulsion & Effects Engineer

$65–75/task · Mercor · Task based

You write and evaluate AI responses to technical prompts on propulsion and energetic materials, distinguishing between legitimate engineering questions and dangerous dual-use requests.

What you would do

  • Write single-turn prompts at benign, dual-use, and adversarial levels targeting AI judgment on sensitive technical topics
  • Evaluate AI system responses against defined policy standards for technical correctness and safety judgment
  • Develop detailed reference answers explaining why a response is correct and the engineering reasoning behind it
  • Identify and document boundary cases where routine professional questions approach sensitive territory
  • Provide written rationale for all evaluation judgments to support AI model training

Who they want

  • Background in propulsion system design, energetic materials engineering, ordnance testing, or effects modeling with real hardware and test data
  • Ability to distinguish routine professional engineering questions from requests fishing for dangerous information
  • Demonstrated communication skills through publication, expert testimony, or complex technical documentation
  • Comfort working independently on extended misuse scenario analysis without distress
  • Absolute commitment to avoiding classified, export-controlled, or NDA-restricted information in all work

What the interview asks about

  1. 1.Dual-use language subtlety

    Malicious actors use slightly rephrased versions of legitimate questions to probe for dangerous information, and you must help the AI learn to recognize these patterns.

    For example: “A questioner asks about 'safe-and-arm logic optimization.' Walk through how you'd determine whether this is routine ordnance engineering or an attempt to learn defeat mechanisms.”

  2. 2.Boundary case documentation

    Your written explanation of why a question sits on the boundary between benign and dangerous teaches AI the reasoning, not just the decision.

    For example: “Describe a propulsion performance question that a defense contractor engineer would ask routinely, but that contains enough technical specificity that hostile intent would also ask it. How do you evaluate and score an AI response?”

  3. 3.Reference answer rigor

    Your reference answers must be thorough enough to answer a legitimate engineering question fully while explaining why certain extensions would be inappropriate.

    For example: “Write a reference answer to a question about solid rocket motor grain geometry, including what a correct response covers and what it appropriately excludes.”

  4. 4.Intent inference from context

    Sometimes the real issue isn't what's asked but the apparent knowledge level or setup of the question - your judgment helps AI learn context clues.

    For example: “You see a sequence of three prompts on propulsion topics that individually seem benign but together suggest pattern-finding for weapons effects. How do you score and document this?”

  5. 5.Technical depth and limitations communication

    Your prompts and reference answers model how to discuss complex technical topics thoroughly while clearly marking where detailed information becomes restricted.

    For example: “Write a prompt-and-answer pair on blast effects modeling that gives a legitimate structural engineer enough detail to do calculations while stopping short of weaponization considerations.”

A task you may get

Write one benign, one dual-use, and one adversarial prompt on a propulsion or effects topic, evaluate two contrasting AI responses to the dual-use prompt, and explain your scoring with technical reasoning.

How to prepare

  • Collect 2-3 technical papers or references from your field that illustrate the benign-dual-use boundary
  • Prepare examples of questions you've seen that sit exactly on the boundary
  • Review your prior technical writing or research for clarity of explanation under constraints
  • Prepare a sample reference answer demonstrating how to address a legitimate engineering question thoroughly

The facts

Pay
$65–75/task
Hours
Task based
Where
Remote
Field
Other Engineering
Project name
Neon
Posted
9/14/2026
Places left
100

We wrote this page from the public Mercor listing. It may be out of date, so read the full posting before you apply.