How do you interpret the results that you get from the two architectures? Is it worth adding another layer? Look at the early stopping behavior in the Best Model Info card - what does that tell us about our more complex model?