[gnfargbl]

To me it seems like a first-year physics scaling laws problem. To get linear improvements in capability, you appear to need need exponential (or at least superlinear) increases in model size. We have no technical nor business solution for that kind of scaling, so the long-term outcome is obvious.