[eggdaft]

Fine tuning can take a fraction of the resources required for training, so I think the original point stands.

[nightski]

Maybe in isolation when only considering a single fine tune. But if you look at it in aggregate I am not so sure.