@dias.social A great article by Pedro Dias. While other articles and commentary about "model collapse" theory usually focus on the long-term degradation of base LLMs during training, Pedro shows how this phenomenon is happening in real-time systems.
theinference.io/p/the-ourobo...
theinference.io
The Ouroboros Learned to Cite Itself
Everyone’s been watching the wrong pipe. The synthetic content is already in the retrieval layer, and the answer engines are laundering it as fact.
