Do they need new training data? I'm pretty sure Midjourney, OpenAI etc. already scraped all of the images in existence long before the well was poisoned. They c

Do they need new training data? I'm pretty sure Midjourney, OpenAI etc. already scraped all of the images in existence long before the well was poisoned. They can use other methods, such as improved tagging, to improve their models.