I think the current "path" of AI will come to a dead end. Until we have AGI, we are just relying on specialised LLMs or other models as components to do various tasks; whether its guessing the next word or picking out objects from an image.

It's almost like we have various functional parts of a brain but not synapsed with some AI-bus. How do you build a performant API for that, which will be as good as a human?