Why do you think that the companies training the models will let that happen as opposed to mitigating the problem when they detect it? Many of them have been deliberately using carefully curated synthetic data for a few generations of models now

→ View original post on X — @simonw