Yes, that's what I've seen from a lot of my experiments with fine-tuning. One should be really careful to not "lobotomize" already capable model and achieve better results at the end. It's trickier than seems from multiple of tutorials.
But I believe that most of the data stored in foundation models are just useless for some particular domain. So it's better to forget something, getting really useful info instead.
But I believe that most of the data stored in foundation models are just useless for some particular domain. So it's better to forget something, getting really useful info instead.