When it comes to generative ai it's not really a content problem, it's actually more of a filtering problem. Whether data was created either partially or fully

When it comes to generative ai it's not really a content problem, it's actually more of a filtering problem. Whether data was created either partially or fully by an AI doesn't actually matter. What matters is the aesthetic quality of the data, as assessed by human tastes, and the diversity in the aesthetic nature. If you can raise the average quality, while maintaining representative diversity of the training data (or fine-tuning), it shouldn't really matter if some of that data comes from generative AI. However this is clearly an unsolved problem. Goodhart's law is a thing and ML is likely to run into similar issues as Search did when attempting to automate aesthetic assessment. You also have issues with AI tools exasperating the content filtering problems by drastically increasing the content to be filtered, as well as distorting the aesthetic diversity. Things that AI is currently already particularly good at will tend to be over-represented in this additional data. I expect large, well-filtered data sets will become closely guarded competitive advantages. I also wonder if we will see Models that get trained/fine-tuned specifically for the task of curating high quality data sets, similar to how we have both content creators and content reviewers/curators as unique roles in human society.