Watched the stream of this on moose and woodchucks.
I guess my big take-away is that in situations like that, increasing the size of the dataset actually probably does improve results.
The problem you run into, and I didn’t hear her address, is that eventually your dataset exceeds what you’ve got to analyze it.
You could build something that’s damn near perfect in differentiating between woodchucks and moose, but does that matter when it takes three years to do the analysis?