AI-generated content is increasingly flooding the web, causing generative AI models to experience a flattening cycle reminiscent of inbreeding, termed 'Habsburg AI'.
The rising tide of AI content complicates efforts to exclude synthetic data from training sets, making it harder for models to maintain their quality.
The phenomenon called 'Model Autophagy Disorder' highlights the risks of AI systems feeding on their self-generated data, leading to degraded model performance.
A recent study illustrated this issue, where repeated AI generations of a simple cooking instruction devolved into incomprehensible gibberish within just a few iterations.
Collection
[
|
...
]