Skip to main content
Skip to content
Menu

Evaluations Intermediate 3 parts Updated

Pitch a safety-eval process for your team

You'll have a 3-slide pitch deck proposing what you'll test before rolling out an AI tool to your team, including which failure modes matter most and how you'll catch them.

What you'll make

Safety-Eval Pitch Deck

  1. Slide 1: The failure modes we're actually worried about
  2. Slide 2: How we'll test for them before deployment
  3. Slide 3: The safety threshold and rollback plan

Upload your slide deck as images or PDF pages

The 3 parts

Each part ends with a check you can see.

  1. Map Your Real Failure Modes

    Identify the specific ways AI could fail in your team's actual workflow context

    Done when

    You have 3-5 concrete failure scenarios with examples of what would go wrong and who would be affected

  2. Design Deployment-Matching Tests

    Create evaluation methods that mirror your actual deployment environment and catch the failures that matter

    Done when

    You have 2-3 specific tests that use the same inputs, outputs, and constraints as your real deployment

  3. Build the Safety Pitch

    Create a 3-slide presentation that makes the case for your evaluation process to decision-makers

    Done when

    You have a complete 3-slide deck with concrete failure modes, specific tests, and a clear safety threshold

    Capstone: you put together your Safety-Eval Pitch Deck.

Where this helps

An elective: a single lesson that sits outside rtwo's job certificates and doesn't count toward one.

About this lesson

What you'll be able to do

  • Present a 3-slide safety-eval pitch that identifies your team's top failure modes and tests
  • Know which eval methods work for your use case vs which ones give false confidence
  • Have a tested evaluation process you can run before deploying any AI tool

Who it's for

  • Team leads considering AI tools who need to make the safety case to leadership
  • Anyone who's been asked 'how will we know if this AI thing goes wrong' and wants a real answer

What it won't cover

  • Academic evaluation benchmarks or research methodology
  • Building custom evaluation datasets from scratch
  • Statistical significance testing or confidence interval calculations

Skills you'll practise

  • Failure mode identification from real use cases
  • Evaluation method selection based on deployment constraints
  • Stakeholder communication about AI safety