Bixby gets better at generalizing as you train it by providing more variations of the same utterance. We recommend atleast 3 utterance trainings for each unique utterance.
Additionally, if you have an utterance that must work, ensure that it is one of your training utterances. Eventually, as you build a more robust set of trained utterances, Bixby will be able to address novel utterances more reliably.
Let’s consider training for the following utterance:
"Who is Mario?"
In order to teach Bixby that the goal of the “Who is [Character Name]?” utterance is the FindCharacter action, we would need to provide at least three utterances where we train the plan.
Let's teach Bixby how to handle this utterance and others like it.
"Who is Mario?"
"Who is Luigi?"
As you can see in this utterance training, Bixby still needs us to provide a goal and highlight the proper values in order to generate a plan. This is because Bixby needs more than one utterance to be trained before it can see the pattern.
Now let's attempt a third utterance training.
"Who is Yoshi?"
Here we see that the earlier utterance trainings have given Bixby enough information in order to suggest what it thinks is the correct path; And its 100% correct!
At this point, we can be reasonably sure that Bixby has learned how to handle this type of utterance. Bixby should not have any issues with generating the correct course of action for any similar utterances.
While Bixby may not require a set of three examples in order to learn a simple utterance such as the example above, it will be essential to provide multiple utterance examples when you are teaching Bixby to learn complex utterances.
Please sign in to leave a comment.