> I worry when I talk to my friends at Google, OpenAI, and other well-funded institutions that their easy access to massive resources is stifling their creativi

> I worry when I talk to my friends at Google, OpenAI, and other well-funded institutions that their easy access to massive resources is stifling their creativity. Why do things smart when you can just throw more resources at them? I'm not convinced that this is a real problem. The big tech companies have way more compute than anyone else, so they should do the large-scale experiments that no one else is able to do. It's true that you don't have to be particularly creative to come up with a lot of these experiments, but nevertheless they should be done and there is a lot of scientific value there. I absolutely agree that more attention needs to be paid to smaller institutions since they're just as likely as anyone to come up with the next great idea. Double blind conference submissions help with this to an extent, but they're not a panacea. If you see an anonymous paper that has performed a thousand Imagenet experiments, is there really any doubt where it came from? And similarly, if they didn't perform a bunch of Imagenet experiments, is there really any doubt that it didn't come from one of the big players? So now you can mask a bias against small institutions as a subtler "you didn't perform enough Imagenet experiments" excuse. (Incidentally, this was the main reason that the paper by Smith & Topin was rejected from ICLR [1].) EDIT: Full disclosure, I'm currently working on a set of not-so-creative (but still IMO important) experiments that use a large amount of compute at Google. So I have some bias here. :-) [1]: https://openreview.net/forum?id=H1A5ztj3b https://openreview.net/forum?id=H1A5ztj3b