Can't you address this with adversarial training items that build upon each other?

I will be doing some testing of this against the base davinci model over the next few weeks.