I don't understand the economics of that, is it just because compute is very constrained and they are buying everything they can, even suboptimal machines with a lower performance per $?
Or is it just an experiment and anyway 10k macs are just a blip?
I hope this means OpenAI might become a contributor to Asahi Linux, lest their hardware investment become depreciated due to planned obsolecense by Apple.
AI-junkie: Claude, please write an AI infra article about OpenAI buying Macs!
Please use a non-beige background for once!
AI-royalty: I'm sorry, non-beige requires the full token price!
Good luck with that. Managing zillions of Macs has always been difficult, clients and servers. Stanford tried to build an Xserve cluster and failed miserably at it because the OS wasn't conducive to it, nor the hardware at the time either. :( It's not impossible but requires a lot of messy custom, fragile tooling.. and that's expensive to support.
Something else they want to price us out of?
I don't understand the economics of that, is it just because compute is very constrained and they are buying everything they can, even suboptimal machines with a lower performance per $? Or is it just an experiment and anyway 10k macs are just a blip?
I think they just want Apple devices to train their computer-use models in a realistic environment.
What realistic environment requires 10,000 Macs?
I mean an agent probably only uses one mac at a time. But they're obviously training tons of agents concurrently.
Aren't they used doing this so they can generate training data for using Macs via the UI with computer use?
The tinfoil hat answer is they want to ensure fewer people can run local LLMs instead of paying OpenAI for their models.
But the monthly subscription isn't that expensive?
(demon rises up. "muahahaha, you don't pay here with money")
Just to make sure we’re gonna do opening and run locally?
I hope this means OpenAI might become a contributor to Asahi Linux, lest their hardware investment become depreciated due to planned obsolecense by Apple.
I highly doubt that a purchase this large did not come with contracts to get software working how OpenAI wants on the apple software hardware stack.
Apple wants this knowledge for their software stack too so it's not going out back into Asahi
Related:
Apple Caught Off Guard by AI Demand for Mac Mini and Mac Studio
https://news.ycombinator.com/item?id=49508982
Good luck with that. Managing zillions of Macs has always been difficult, clients and servers. Stanford tried to build an Xserve cluster and failed miserably at it because the OS wasn't conducive to it, nor the hardware at the time either. :( It's not impossible but requires a lot of messy custom, fragile tooling.. and that's expensive to support.
Sounds like a job for a state of the art model...
[dead]