@jonny I would argue that this is a common feature of these models. The other day I asked for a translation and it gave me the exact opposite of what was written (swapped secondary vs primary), presumably because the latter was more common in training data than the former.