We're at a weird juncture where people are heralding language models as while simultaneously acknowledging they're 99% regurgitating existing text - aren't we just outsourcing our problems to machines that can't actually understand what we're asking?