I don't think it's just a gate-keeping tool. (Though it certainly has that effect.) We're rapidly approaching a point where, like for many other fields before i

I don't think it's just a gate-keeping tool. (Though it certainly has that effect.) We're rapidly approaching a point where, like for many other fields before it, there's just not much interesting amateur work to be done on large models anymore. You need to be affiliated with some sort of institution to get access to the necessary resources. Maybe theoretical work such as developing fundamentally new neural network architectures that don't need to be proven out with a 7+ figure budget for training the model? But that would be more of a labor of love than a career strategy.