So we have LLMs that can represent the language part of the brain. Robotics and self-driving is emulating the spatial part of the brain. The image and video generators are the visual processing part of the brain. Independently these are components of a brain but they don’t equal intelligence. These are all things that animals are capable of in various forms. Sometimes I space out when I’m driving the car - it takes up a small fraction of my brain.

There’s a really big thing missing before AGI can be developed: reasoning and self learning. So we would need a model that knows its limits (the best LLMs don’t seem to), asks for more information, and then improves its own model. My suggestion for this would be to make a new AI that tries to teach itself mathematics. Start with numbers and representations of numbers, then addition, multiplication, and so on. But the key is the initial training has to be the building blocks only and the rest it has to learn itself by asking human teachers. When you get into higher math and it makes novel inferences, I think that will be the first sign of intelligence. Then you can get into physics and language and other stuff and it’s still a long path from there to AGI.

Anybody have examples of self-improving models? Following that might indicate whether we may have AGI by 2030. I’m skeptical.

Nope.