The end goal isn’t a system that knows everything. I’d argue a system that can learn from context but has very little knowledge about the world would be much mo

The end goal isn’t a system that knows everything. I’d argue a system that can learn from context but has very little knowledge about the world would be much more interesting than our current LLMs. GPTs have a small spark of higher level reasoning. Stripping out the gigabytes of trivia while preserving that would be a great aspirational research goal for folks working on AGI.