Think Together · Start with a Short Story
The girl copies the same apple photo ten thousand times. The robot asks, “Is that really more helpful than pictures of ten different apples?”
Lay Out the Process
- QuantityHow many examples are there?
- QualityAre labels and content reliable?
- CoverageAre different situations included?
- EvaluationDoes it improve on new examples?
Data value depends on quality, coverage, and relevance to the task, as well as quantity.
Large amounts of repeated or incorrect data can strengthen bias and raise costs without adding useful information.
Adding suitable, representative examples of cases where a model often makes mistakes can be more helpful than repeating examples blindly.
Now You Try
If every dog photo shows the same dog, does having many photos guarantee accuracy on an unfamiliar dog?
Ready? Open this to see the explanation
No. The examples may not cover different appearances and backgrounds. Check with varied new examples.
One Idea to Take Away
Data value depends on quality, coverage, and relevance to the task, as well as quantity.
Try covering the diagram above and explaining each step to someone in your family. If one step is hard to explain, return to that card and look again.