I had a great chat with a developer from Vicarious last night. We talked about how solving for creative founders was super critical, since the unique training data-set required needs to be obtained creatively.
But there was another dimension that I had not anticipated, which is what happens if the number of data-sets wasn't as high as needed. If that's the case, one creative solution would be to reduce the variation in data so that the algorithm is more efficiently trained.
We had other discussions about constraints to the indexing of early stage AI/ML startups, such as cloud computing usage and the knowledge gap. More for the next time.