We get the same complaint in ML - big companies have more money for compute so they can try more idea variations, discard the bad ones and stake a claim on the

We get the same complaint in ML - big companies have more money for compute so they can try more idea variations, discard the bad ones and stake a claim on the good ones. Or even on the same architecture they can train more models and by pure chance get one that is %0.1 better than previous SOTA ("random seed optimization").