It is not really possible to lose 90% of one's brain and do well on certain narrow tasks. If the tasks truly were so narrow, you would be better off training a small model just for them from scratch.