One area that OpenAI did not comment on, was the likelihood that the AI-generated content here was itself used in the training data for a later model.
Let's say the OpenAI engineers are working on ChatGPT 5 and spent last month scraping Teorater and X/Twitter, where this material ended up. How does OpenAI know that the new model is not poisoned?
This isn't just OpenAI's problem of course. Anyone training on the open Internet now has this problem.