I really enjoyed this part of the plan: The profiling mode would generate a random seed, generate the code, compile the code, run a program that measures the p

I really enjoyed this part of the plan: The profiling mode would generate a random seed, generate the code, compile the code, run a program that measures the performance of the code, and compares the execution time to the previously-known best execution time. Then we would just let it run until we run out of development time, take the best seed the program output, update the build system so that the new seed is used for production builds, and say we’re done. We could call our ridiculous compiler the “super dumb compiler” and hope that people confuse it for a superoptimizing compiler. To scale this incredibly inefficient optimization strategy, we could mint a new cryptocurrency, Superdumbcompilercoin, where we trick people into running the super dumb compiler on their computers, cross-certifying each others’ results, and sending us increasingly better optimization seeds in return for plausible delusions of becoming rich.