Job posts for AI training use a lot of technical words. You don't need to understand AI to do the work, but these terms help you read listings and project guidelines. Each term has a one-sentence meaning and, where useful, a finance example.
The 12 words you'll see most
Term
What it means
Finance example
AI training
Paid work where experts check and improve the answers of AI systems.
Reviewing how an AI explains goodwill impairment.
Model or LLM
The AI system itself. LLM stands for "large language model", the engine behind tools like ChatGPT, Claude or Gemini.
"The model" is the AI you're testing.
Prompt
The question or instruction given to the AI.
"Build a DCF for this company using these assumptions."
Response or completion
The AI's answer to a prompt.
The DCF the AI produced.
Rubric
A scoring checklist of what a good answer must contain. It works like a marking scheme for an exam.
"Uses mid-year convention", "reconciles enterprise value to equity value" and "states WACC inputs".
Golden answer or reference answer
A model answer written by an expert, used as the benchmark.
Your fully worked solution to an LBO case.
Evaluation (eval)
A test of how well the AI performs a task.
A set of 100 accounting questions with expert answers.
Benchmark
A standard set of evals used to compare AI systems.
Mercor's APEX benchmark includes investment banking tasks.
RLHF
"Reinforcement learning from human feedback": the AI improves by learning from experts' ratings of its answers.
You rate two answers on deferred tax; the AI learns which is better.
SFT
"Supervised fine-tuning": training the AI on examples written by experts.
You write 20 model answers on IFRS 9; the AI learns from them.
Red teaming
Deliberately trying to make the AI give wrong or harmful answers, so they can be fixed.
Writing a tricky question that makes the AI confuse cash and accrual accounting.
Annotation or labeling
Adding labels or notes to data. Mostly entry-level work; expert work is usually reviewing and writing.
Tagging lines in a financial statement as "revenue" or "cost".
Words about how the work is organized
Term
What it means
Project or pipeline
A specific piece of work for one AI company, with its own rate, tasks and hours. It is what you are hired for. micro1 calls them "pipelines".
Task
One unit of work, for example one prompt to review.
Queue
The list of tasks waiting to be done. An empty queue means there is no work right now.
Pod or team
A small group of experts working on the same project, often with a lead.
Reviewer or QA
Someone (often another expert) who checks your work.
Guidelines
The project's instructions. Follow them exactly.
Work trial
A short, usually paid test project before you're hired, for example at Mercor.
Assessment
A skills test. It is often unpaid, and sometimes you get one attempt only.
Offboarding
Leaving a project, because it ended or because of quality or rule issues.
Paused
A project is on hold. You usually can't log hours.
Per-task pay
You're paid a fixed amount per approved task, not per hour.
Time tracker
Software such as Insightful or Hubstaff that records your hours and may take screenshots.
Independent contractor
You work for yourself, not as an employee. You get no paid leave and you handle your own taxes.
W-8BEN
A US tax form non-US contractors sign to confirm they're not US taxpayers.
KYC or ID verification
"Know your customer": checking your identity with a passport or ID.
Words you'll see in finance job posts
Phrase
What it usually means
"Domain expert"
Someone with real professional experience in a field. That's you.
"Top-tier firm"
Big Four, a bulge-bracket or elite boutique bank, a top private equity fund or a top-three strategy consultancy. It helps but isn't always required.
"Build from scratch"
You create a model or case yourself instead of reviewing one. It pays more and takes longer.
"Stress-test the model"
Find where the AI fails.
"Rationale"
A written explanation of why you scored an answer the way you did.