I suppose we’re about to find out if these things (LLMs) are any good at extrapolation. I expect not really as they’re effectively just interpolation machines.