Reflection AI’s definition of being “open” seems to center on access rather than development, similar to strategies from Meta with Llama or Mistral. Laskin said Reflection AI would release model weights — the core parameters that determine how an AI system works — for public use while largely keeping datasets and full training pipelines proprietary.
Better than nothing, but I'd love to see open datasets and pipelines.
Currently this status quo sustains the chipmakers, but they'll be more than willing to invest in your fully open source dataset and pipeline once they run out of datacenter hype: then everyone will be needing compute.
Also, the government having Intel's back means that being outcompeted is meaningless and you just have to be too big to fail. Nvidia won't ever go down: it'll be national interest, so they can leverage being a government backed entity now, just like JPM.
Better than nothing, but I'd love to see open datasets and pipelines.
Currently this status quo sustains the chipmakers, but they'll be more than willing to invest in your fully open source dataset and pipeline once they run out of datacenter hype: then everyone will be needing compute.
One step at a time?
Good point! It makes me wonder if that's what happened with the internet - over/mal-investment so severe it yields a near-public good.
Also, at some point, there will be so many companies that are "open" that it won't be competitive advantage.
I just enjoy being impatient.
Possibly, yes.
Also, the government having Intel's back means that being outcompeted is meaningless and you just have to be too big to fail. Nvidia won't ever go down: it'll be national interest, so they can leverage being a government backed entity now, just like JPM.