You mentioned Grid Former, i found a paper describing it (Grid Former: Towards Accurate Table Structure Recognition via Grid Prediction). How did you implemented it?
My opinion is that the work they are performing it's not the point of them being there. It's about testing the limits of the exploitations of human beings.
I tried this with LeChat (mistral) and ChatGPT 3.5 (free) and they start to respond to "something" following the style but... without any question asked.
We are unconsciously (pun intended) implementing how brains work both in dream and wake states.
Can't wait until we add some kind of (lossless) memory to this models.