Comment by AnimalMuppet

It may be a positive feedback loop - whichever language the AIs write more of becomes more predominant in future training data.

But if AIs write bugs in that language, they also poison the well...