AI companies should be donating large sums of money to Wikipedia and other such sites to keep them healthy. Without good sources, we’re going to have AI training off AI slop.
They should be, yes, but they won't. We already know this from the way those same companies have treated open source projects that they depend on.
One thing that I would really like to see is some kind of hefty tax on any kind of income derived from models trained on Wikipedia. Basically, make it legal to train, to share weights etc freely, and hosting them locally. But the moment you start charging people for subscription, the society should start charging you to maintain the commons that you are profiting from.
(This likely goes for more than Wikipedia, but that case is especially simple since there's a single legal entity that could be given the money.)
AI companies should be donating large sums of money to Wikipedia and other such sites to keep them healthy. Without good sources, we’re going to have AI training off AI slop.