See normally when an algorithm requires The Entire Global Supply Of Ram, a software engineer would consider that not a good algorithm, and would try and make a better one instead of optimising the entire planet for paperclip manufacturing.

Unfortunately we accidentally gave a bad software engineer too much money

@bri7 after learning even the basics of why neural nets produce surprisingly good results with limited resources, it makes no sense to bother scaling them, which led me to developing a theory that eating the resources and fishing for subsidies was the actual design goal of the architecture,
@bri7 (i have another theory that explains the seemingly braindead design choice, and that one relies on purposeful monopoly and internalized exceptionalism based on racism, and the basis of that idea is already debunking itself in real-time)