Do LLMs suffer from catastrophic interference? A SWE is expected to learn continuously as the dujour stack constantly changes under our feet. This can be brute

Do LLMs suffer from catastrophic interference? A SWE is expected to learn continuously as the dujour stack constantly changes under our feet. This can be brute force mitigated by epochal retraining from ground up but that will cost money. So the cost equation must include a periodic 'rebuild of the dev team'.