Model Cards and Datasheets for Datasets Quiz

Answers and explanations appear as you take the quiz.

1. What is the main purpose of a model card?
   • To advertise a model's best benchmark score
   • To document a model's intended use, evaluation, performance across conditions and limitations
   • To store the model's weights
   • To license a model for commercial sale
2. Who led the paper that proposed model cards?
   • Geoffrey Hinton
   • Fei-Fei Li
   • Margaret Mitchell
   • Yann LeCun
3. What is disaggregated evaluation?
   • Reporting performance separately for relevant subgroups and conditions
   • Averaging all test results into one score
   • Removing evaluation data from the model
   • Testing only on synthetic data
4. Datasheets for Datasets borrows its idea from which field?
   • Pharmaceutical labeling
   • Restaurant menus
   • Financial auditing
   • Electronics component datasheets
5. How does a system card differ from a basic model card?
   • It covers only the training data
   • It documents a whole deployed system, including safety testing and mitigations
   • It is legally required in the US
   • It contains the source code
6. On Hugging Face, where does a model card live?
   • In a separate PDF
   • In the model weights file
   • In the repository's README.md with a YAML metadata header
   • In a private database only staff can see
7. Which EU AI Act requirement relates to general-purpose AI model providers?
   • Publishing a summary of training content using a Commission template
   • Releasing all weights publicly
   • Obtaining a CE mark for every chatbot reply
   • Registering each user
8. Why is it a misconception that a model card proves a model is safe?
   • Model cards are always written by regulators
   • Model cards cannot include metrics
   • Model cards are only for image models
   • A model card is a disclosure, not a certification, and it can be incomplete or outdated