Мрежа је медиј

Some have started to refer to this type of content as “AI slime” (your feeds are covered in it).
What’s worse than this flood of AI-generated slime taking over our social media feeds is that many people, especially in older generations, don’t always appear to realize that the images or bits of content are AI-generated. Based on the comments posted, many seem to believe the images are real photos or artwork created by actual human beings.

The idea of using AI-generated data to train other AIs is not new. It has clear benefits, such as reducing the cost and time spent collecting and annotating real-world data. However, according to a recent study covered by ArsTechnica, there’s evidence suggesting that machine learning models trained this way may be getting worse over time.
In the world of machine learning, we call this phenomenon ‘model collapse.’ It occurs when a model trained on synthetic data starts generating increasingly less diverse and more repetitive outputs. The AI essentially becomes an echo chamber, repeating the same patterns without the capacity to generate novel responses.

OPT-125m -- was fine-tuned using only artificial model data in one scenario and a mixture of human-generated data and artificial data in another. Researchers found that although model performance degraded over time, some level of learning was possible with generated data. Still, the example given in the study showed several outputs from OPT-125m responding to prompts about medieval architecture in which, by the fourth generation, the model was outputting completely unrelated text about jackrabbits.