Generative AI Might Be On The Verge Of Self Destruction
Model Collapse is a genuine risk.
While AI is nowhere near as capable as many claim, it is nonetheless being integrated into our everyday lives. Particularly generative AI. For example, ChatGPT seems to now be integrated into every operating system known to man. Similar generative tools are becoming widespread in numerous industries, even if they don’t positively impact them. What’s more, AI-generated content is taking up more and more of the corners of the internet we call home. Whether we like it or not, every inch of our lives is being touched by this technology. So, the fact that new research suggests generative AI models could collapse into a state of meaningless babble in just a few years is deeply worrying. Let me explain.
To understand this issue, we first have to understand how AI works. So, let’s quickly recap that. AI works by training an artificial neural network on a vast amount of well-labeled data. This neural network finds trends in this data, enabling it to recognise similar patterns in other data, predict how other datasets might evolve, or extrapolate these trends to create something new but derivative of the original dataset.
Generative AI falls into the extrapolation camp of AI. Take ChatGPT4, for example. It was “trained” on 570 GB of carefully selected…
