https://aiindex.stanford.edu/wp-content/uploads/2024/04/HAI_AI-Index-Report-2024.pdf

Blogger at ribbonfarm.com, running summerofprotocols.com, host of /gloom ☀️

Figure 1.3.9 demonstrates the process of model collapse in a variational autoencoder (VAE) model, a widely used generative Al architecture.
With each subsequent generation trained on additional synthetic data, the model produces an increasingly limited set of outputs.”

“For instance, a team of British and Canadian researchers discovered that models predominantly trained on synthetic data experience model collapse, a phenomenon where, over time, they lose the ability to remember true underlying data distributions and start producing a narrow range of outputs…