The article compares the contamination of steel after World War II due to nuclear fallout to today's saturation of AI-generated content on the internet. Following the rise of AI models like ChatGPT in late 2022, the digital space is filled with text synthesized by machines rather than humans. This raises concerns about originality and quality, as AI models trained on previous AI-produced data may lose their creative edge, creating a cycle of dilution known as model collapse. Consequently, pre-2022 human-generated content becomes increasingly valuable for grounding AI and society in a shared reality.
What we are witnessing today is a similar effect to the contamination of steel post-1945, where AI-generated content is overwhelming our digital landscape.
If today’s AI is trained on yesterday’s AI-generated text, it risks folding in on itself, leading to model collapse and loss of originality.
Collection
[
|
...
]