One thing I found about intelligent LLM applications is that Intelligence primarily
resides not in the application interface itself, but within the model. The role of the
application interface is to change the way humans and intelligence interface.
resides not in the application interface itself, but within the model. The role of the
application interface is to change the way humans and intelligence interface.
By native, I mean the intelligence should be closer to the operating system, ideally
native inside the operating system.
The design of interface with powerful AI should be the primary concern at the
application level. What makes Cursor successful is that the interface utilizes
intelligent agent to reduce human's barrier to understand and write code. Essentially,
this is achieved through the innovation of interface. For example, Cursor, being an
app rather than a browser-based tool, can directly manipulate the underlying OS.
The design of interface with powerful AI should be the primary concern at the
application level. What makes Cursor successful is that the interface utilizes
intelligent agent to reduce human's barrier to understand and write code. Essentially,
this is achieved through the innovation of interface. For example, Cursor, being an
app rather than a browser-based tool, can directly manipulate the underlying OS.
The fact that interface being native allows intelligent model to better operate
the underlying operating system. Whereas Smartness of interface is less important
because artificially set smartness in the interface could hinder the model's inherent
capabilities. This is also aligned with 'Bitter lesson' that long term only most native UI
would scale well with increasingly capable intelligence.