40 to a trillion pieces, depending on your goals.
The answer to this question will change for everyone - the more data you have to begin with, the more accurate your AI blocks will be. Read on to see an example of this answer in action, and learn how to set your own goals for the amounts of data you might need.
Did you know: Levity now offers Templates for some use cases, which means you can theoretically get started without any of your own data.
The system asks for a minimum amount of data to get started. Below is the accuracy of our AI block with 20 photos of each breed uploaded (a total of 40 pieces of data overall):
As you can see, the model isn't particularly sure what type of cat Franny might be. Let's try that again with 100 photos of each breed type:
With a solid sampling of 100 photos of tuxedo cats, we improved our model's performance dramatically. Let's try again on our improved model with a different camera angle of Franny:
As you can see, the model lost a little confidence as the photo isn't framed as well. We can keep training it with the many cat photos we've taken over the years to improve it further.
If you are just getting started, about 100 pictures, documents, or text examples per class are already enough to deliver solid results. Just keep in mind - the more specific you want to get, the more training data you'll likely need.
We can also help you to gather more data if needed - just reach out to us on our in-app chat.