It's an interesting idea, to allow more natural entry as data entry is one of the highest friction parts of this. But from my perspective the examples used to show and sell this are very bad.

User: "Paella"

App: A huge set of undisplayed assumptions on ingredients and serving size resulting in a hilariously low estimate of calories compared to what any restaurant serving of Paella would hold along with the typical wordy generic AI text output that doesn't help anyone.

User: "Chicken Caesar Salad"

App: Same as above. Presumably someone eating a Caesar salad already knows what they ate and doesn't need to be told it has chicken and lettuce in it.

The exercise examples are "a rough estimate" and doesn't appear to include any user-specific data such as speed or heart rate.

Is it possible to show examples that would fit better with how a real user might use this. Is it capable of getting a good number for something like "qboba burrito bowl, white rice, black beans, chicken..." and so on? Can it remember things so a user could say "my usual burrito bowl"?

And can the inane and utterly useless AI text be turned off?