Well, if openai open sources all their models, what value is there left that is unique to openai? Training data? But that is only valuable if they invent a better way to make it into an “AI”.
Hosting a model of that size requires ~800GB of VRAM. Even if they release their models, it wouldn’t make them obsolete since most people and many companies couldn’t host it either way.
Most companies that have open source products make their money in providing support for it to other companies, so theoretically OpenAI could do that (in fact top level AI developers are pretty rare so they could charge way more than other companies do for consulting) but that wouldn’t be a hundreds-of-billions-of-dollars business plan because it’s much too sensible.
Well, if openai open sources all their models, what value is there left that is unique to openai? Training data? But that is only valuable if they invent a better way to make it into an “AI”.
Hosting a model of that size requires ~800GB of VRAM. Even if they release their models, it wouldn’t make them obsolete since most people and many companies couldn’t host it either way.
OpenAI started as a non-profit, so it could be 100% open and still be truthful to it’s original intentions.
Most companies that have open source products make their money in providing support for it to other companies, so theoretically OpenAI could do that (in fact top level AI developers are pretty rare so they could charge way more than other companies do for consulting) but that wouldn’t be a hundreds-of-billions-of-dollars business plan because it’s much too sensible.