How intelligent will large-scale models become in the foreseeable future? The intelligence of large-scale models is approximately equal to Moore's Law of computing power plus the scaling law of models. Moore's Law of computing power increases fourfold every 18 months. One explanation of the scaling law is that existing training data can support two such cycles. So, in three years, in 2029, large-scale models will be 16 times smarter than they are now. Do you find this surprising or alarming?