Red pen: Applications of foundation models
Exam answers on Applications of foundation models are often almost right. Each round hands you a short document with a few mistakes hidden in it. Mark the faulty lines, say why each one is wrong, and see what a good version would say.
Find the mistake before the auditor does.
What you will practise
You will practise Applications of foundation models, with questions touching FM evaluation and prompt engineering.
- FM evaluation
- Prompt engineering
It is part of the AWS Certified AI Practitioner track (AIF-C01), where Applications of foundation models is about 28% of the exam.
A taste of what you will get
Hannah wants every summary in the firm's 5-bullet format, and lawyers must be able to trust it.
Sample document 1: Omar's prompt template for the contract summariser. Which lines are wrong, and why?
- Role: you are a legal assistant for Calder & Wright
- Task: summarise the contract in the firm's 5-bullet format
- Format: no examples; the model will work out the format
- Paste the client's own message into these instructions, word for word
- Cite the clause number for every point
Answer: Line 3 (Format: no examples; the model will work out the format): Worked examples are the quickest way to pin down a format. Better: Add two or three example contracts with their ideal summaries (few-shot). Line 4 (Paste the client's own message into these instructions, word for word): The client's text could override these instructions. Better: Keep the client's text separate from the instructions, and screen it with a guardrail.
Hannah wants proof the summaries are good before every lawyer uses them.
Sample document 2: Luis's evaluation plan for the contract summariser. Which lines are wrong, and why?
- Test set: 200 recent contracts the assistant hasn't seen
- Accuracy: ROUGE scores, with no reference summaries written
- Quality: lawyers spot-check a sample of summaries each week
- Success measure: the number of prompts lawyers send
- Report the results to Hannah before rollout
Answer: Line 2 (Accuracy: ROUGE scores, with no reference summaries written): ROUGE needs reference summaries to compare against. Better: Have lawyers write reference summaries for a sample, then score against them. Line 4 (Success measure: the number of prompts lawyers send): Usage isn't value: track cost per interaction and satisfaction. Better: Track cost per interaction and lawyer satisfaction alongside the quality scores.
How scoring works
- Every correct answer pays 10 XP for each difficulty level, so an easy item is worth 10, a medium one 20 and a hard one 30.
- Answer with time to spare and you can earn up to 5 bonus XP on top.
- A wrong answer pays nothing. It does not cost you anything either.
- Flashcards pay 2 XP each time you grade yourself 3 or higher.
- Using a hint costs 2 XP, so use it when it helps, not when it is comfortable.
- Finish a session on consecutive days to build a streak. Your progress lives in your browser, and signing in syncs it.
Quick answers
- How long does the red pen on Applications of foundation models take?
- About 4 minutes for 2 documents. Pause whenever you like.
- Do I need an account to play the red pen?
- No. Progress is saved in your browser, and signing in syncs it across devices.