Until one day an outsider finds a new approach for LLMs that vastly reduces the computational complexity.

And then we’ll realize we wasted an entire Apollo space program to build an over-complicated autocompleter.