Reactions from ranked influencers
3 posts
Why is this framed as if the argument is that open source should eliminate closed source? These are not mutually exclusive architectures. The closed source providers are doing just great even amidst growing open source options. Secondly, it’s definitely not obvious that open source can’t sustainably be funded by the profits generated from inference and surrounding services. Nvidia, for instance, has extreme financial incentive -without any need from the government- to make sure that there are open source frontier options. AWS operating margins are in the 30-40% range, so they can make plenty of money running infra for models, which of course would flow back into model R&D. Right now those dollars don’t go into US OSS because the dollars are best used by closed model labs, but that doesn’t mean the alternative isn’t financially viable.
Ok fair point, and I totally buy that that’s a different environment than today, and I can see some of the risks that are being presented. My guess is, in response: 1. OAI would more quickly vertically integrate on inference (which you’re already doing), and make money running your models and make sure no one can run the open source stuff cheaper. 2. You’d make plenty of money being the secure enterprise provider of your models, which banks / pharma / etc all need anyway even with open weights. You’d probably get into post training infra and services, etc. 3. It’s not obvious much changes about the end-user product or b2b subscription revenue from the app layer since that’s largely about brand + model anyway. But nevertheless, don’t you have to prepare for some version of that world even without open weights?
@levie @jon_stokes yes they’re doing great because open source is still quite far behind if Sol were distilled and open sourced the day it finished training it would be a different situation