OpenAI CEO Sam Altman said in a podcast interview that "it's quite terrible" for people to have RAM in their own PCs when it could go into AI data centers.
Only a moron would think OpenAI was ever turning profitable.
There were two possible ways,
Subscription prices rose to a point no one would be interested to pay.
Inference got so cheap, that 98% of all use cases could be run by open weight models on local machines
In both situations OpenAi and Anthropic lost.
The bull case has been a bear case since the beginning. Both companies will most likely stay alive, but will eventually have their costs turned down as they’re going to be bought out by some of the mastodonts.
The best possible scenario for either of them, is if a government want a frontier lab and decides to buy one already functioning. Rather than having their universities start from scratch
Maybe, maybe not. Phones are getting more optimized for AI and have decent RAM, so the on device models may not be great at agentic problem solving, but they’ll probably meet most people’s AI needs
Plus, any reasonable gaming computer or MacBook can run models… Maybe a bit slowly and maybe just the compressed 8gb models, but they can pretty much all do it to some degree
So I think we’ll be in a place where there’s little or no upfront cost after a couple more key advancements
Sounds like Sammy sees what way the winds are blowing, and is scared
Models are getting a lot more efficient, and the billion dollar models are going to have to compete with something basically free to run
Only a moron would think OpenAI was ever turning profitable.
There were two possible ways,
Subscription prices rose to a point no one would be interested to pay.
Inference got so cheap, that 98% of all use cases could be run by open weight models on local machines
In both situations OpenAi and Anthropic lost.
The bull case has been a bear case since the beginning. Both companies will most likely stay alive, but will eventually have their costs turned down as they’re going to be bought out by some of the mastodonts.
The best possible scenario for either of them, is if a government want a frontier lab and decides to buy one already functioning. Rather than having their universities start from scratch
They die if we use local models.
The second profit condition would be it got so cheap they could make bank on data collection
Something that has a fixed up front cost to run*
Otherwise, YES!
Depending on your rig there will still be recurring cost (electricity)
Well, yeah… i was just pointing out that “free” wasn’t exactly accurate
I know, I was adding another thing that made it not free.
Maybe, maybe not. Phones are getting more optimized for AI and have decent RAM, so the on device models may not be great at agentic problem solving, but they’ll probably meet most people’s AI needs
Plus, any reasonable gaming computer or MacBook can run models… Maybe a bit slowly and maybe just the compressed 8gb models, but they can pretty much all do it to some degree
So I think we’ll be in a place where there’s little or no upfront cost after a couple more key advancements