I wonder what the outcome will be when new models are trained on AI-generated data. These companies are already running out of quality training data. So when most of the data on the internet is synthetic, will they find ways of separating the signal from the noise, or will all the noise lead to a convergence of performance across all models to something that is much inferior than what we have today?

This tech will make the internet even more unbearable to use, without mentioning its huge potential for abuse. This is far worse than whatever positives it might have, which are still unclear. What a shitshow.