It’s like the opposite of the Lucas Critique: a sufficiently developed model always extrapolates out of sample
alz
alz20.7. klo 23.20
LLMs are obviously not just interpolating data. Sufficiently high-dimensional models are always extrapolating
2,95K