Do current deep learning models have a natural ceiling for intelligence?
Reviewed by Jason Burns, Editorial Steward · Last updated
Not exactly — there is no proven hard ceiling, but there are visible diminishing returns from raw scaling and open architectural limits around planning, memory, and reasoning that current transformer-based models do not clearly overcome. As I. J. Good, put it on the record: "Let an ultraintelligent machine be defined as a machine that can far surpass all the intellectual activities of any man however clever. Since the design of machines is one of these intellectual activities, an ultraintelligent machine could design even better machines; there would then unquestionably be an 'intelligence explosion'."
Source: www.sciencedirect.com/science/article/abs/pii/S0065245808604180