>>105844366
I think you grossly overestimate the compute required to train a model. I also think you grossly underestimate how much compute is wasted undoing the lobotomy / redoing a model's understanding of anatomy.