GPT-4 has been top dog for a while, and it's said to be a network of experts architecture. Even the human brain is similar to some degree with different "sections" of the brain or cortexes being specialized for specific sensory input. It's always made sense to me that a quick way to come to a "good enough" AI is to train many smaller expert AIs and then train a umbrella AI that just delegates parts of the task to different experts, and organizes the output. The next step would obviously be training that umbrella AI to make it's own expert AIs in an unfamiliar supervised fashion.
Is anyone already working on something like this? Do you think it's a worthy avenue of research?
[link] [comments]