> The knowledge and the infra needed to serve these huge models to billions of users reliably seems to me to be a pretty serious moat here that no current open source project can compete with.
I don't quite follow this line of argument. Let's say there's an open-source ML model X with a permissive licence. I think it's not super relevant whether whoever came up with the model graph & weights for X knows how to serve it at scale, as long as someone does. And it seems pretty clear that it's not just Google (and OpenAI) who know how to do this.
Separately, I'm personally more excited about the possibility of running these models on-device rather than at scale in the cloud (for privacy and other reasons).
I don't quite follow this line of argument. Let's say there's an open-source ML model X with a permissive licence. I think it's not super relevant whether whoever came up with the model graph & weights for X knows how to serve it at scale, as long as someone does. And it seems pretty clear that it's not just Google (and OpenAI) who know how to do this.
Separately, I'm personally more excited about the possibility of running these models on-device rather than at scale in the cloud (for privacy and other reasons).