Currently the systems we see solving all kinds of math problems rely on a single large language model. That makes the AI we observe today essentially a machine-amplified version of Type 1 capabilities. I don't think AI is incapable of producing Type 2 capabilities. Instead those will emerge in the future as different large language models interconnect or as massive numbers of AI agents interact with one another. These agents will connect to different large language models and the resulting creativity and imagination could far surpass human levels.

C@C

(view original post →)