I wonder if when multiple LLMs are more widely available if there's money to be made in providing a unified API that falls back to different service providers depending on their status. Multicloud meets AI.
In the medium term I expect AIs to begin maintaining significant state (so that you can teach it contexts and don’t always have to start from scratch) that can’t just be switched to a different provider.
To me it seems more likely that actors with robust systems are enough
Unlikely, as small differences in the models can widely affect the expected responses. If anything, people would pay for a way to freeze the model at a specific version.
I am working on this and seeking a non-technical co-founder. I'm not sure it would make money however.