Really great insights and thank you for sharing this! I wonder if maybe we should go straight to SW 3.0, where the key is removing the dataset problem. Let me explain:

1- dataset will never be complete

2- they will never be sufficient

3- working with dataset we ignore the problem of continual learning and iterate and waste resources

4- we should think about way to grown a mind = embed "concepts" from multi-modal domain into a set that can support add-on apps = tasks

5- work mostly on self-supervision = predict, reconstruct, complete data you do not see

I think this is SW 3.0. What do you think?

More here: https://culurciello.medium.com/drop-the-dataset-b3ed418bd69d, fresh from today!