This will probably sound a lot more natural than sitting there reading a training corpus. Given that advances in AI are tending to eliminate manual feature engineering, predetermined training corpuses may soon be a thing of the past anyway.