- While Robin spokes about Open and Close models, I think he is actually referring to deployment mode- i.e. managed compute mode vs serverless AI mode. Open source model can be deployed in Serverless API mode, hence users can still pay-as-you-go for the tokens instead of paying for the compute infrastructure. - With regards to SFT and Content Safety, they have to be considered by the Application Users, there is no diff between Open and close source model. In that sense, Open source model may not be more expensive than close source model. Can someone clarify? I also have questions about the relevance of Open Source and Close Source for AI models, as the "source" is not able "program". Hence, how relevant is open source vs close source for AI models? Perhaps the methodology of training and models, explainability and observability of the models are more important? Does anyone have comment on this?
Excellent interview. Very informative and useful. But i really want to know his view on the chip embargo on china, whether in the future china can still have access to top AI chips for training and inference. Otherwise china might not have a future in AI anyway, let alone spending time on AI applications. Any thoughts?