Does it work? Evaluating an AI before you trust it
A demo always goes well: the cases are chosen by whoever wants to convince you. What you need to know is what happens with the cases in your own work, including the ugly ones. This course teaches you to measure it in four blocks: deciding what counts as getting it right, building your own exam, choosing who grades it and continuing to test when the system changes.
Private Course
Please sign in to contact responsible
| Responsible | Carlos Pascual |
|---|---|
| Last Update | 2026-10-06 |
| Completion Time | 4 hours 30 minutes |
| Members | 1 |
Share This Course
Share Link
Share on Social Media
Share by Email
Please login to share this Does it work? Evaluating an AI before you trust it by email.
Intermediate
Data, Analytics and AI
-
-
Preview
-
-
Block 1: What getting it right means: the criterion before the grade3Lessons · 1 hr
-
Defining success
-
A demo is not an evaluation
-
Block 1 exam10 xp
-
-
Block 2: The test battery: an exam made by you3Lessons · 1 hr 5 min
-
Building the battery
-
Saving the exam and reading the failures
-
Block 2 exam10 xp
-
-
Block 3: Who grades: people, rules and models3Lessons · 1 hr 10 min
-
Three judges: a person, a rule and a model
-
Is it really better, or is it noise?
-
Block 3 exam10 xp
-
-
Block 4: From the exam to day-to-day: keeping, changing and withdrawing trust3Lessons · 1 hr 10 min
-
Evaluating an agent: the result and the path
-
When the model changes, test again
-
Block 4 exam10 xp
-