Oxford-Cambridge study warns of 'model collapse' risk in AI language models

Researchers from Oxford and Cambridge universities have identified a critical flaw in large language models called 'model collapse'. The phenomenon occurs when generative AI systems repeatedly train on synthetic data they themselves produce, gradually losing the nuanced qualities of original human-generated content. Over time, this process erodes the diversity and authenticity of information these models can generate. Noted investor Michael Burry has echoed these concerns, warning that AI systems risk introducing and amplifying propagation errors in information.