As this article reports, "Machine learning models trained on content created by generative AI will suffer from model collapse, according to a new study" (18 page PDF). Specifically, "What happens when the internet becomes flooded with AI-generated content? That content will eventually be collected and used to train the next iterations of generative models." The article describes in detail how the re-use of AI-generated content corrupts the data. Basically, probable events are over-estimated, and improbable events are underestimated. But if everybody agrees that this is a bad idea (as they appear to) then why would they do it? But also: isn't this what happens with human networks as well: when people just talk among each other, without any new evidence being introduced, they abstract more and more until they end up with just a few polarized points of view? Without openness, networks tend toward model collapse.
Today: 2 Total: 117 [Direct link] [Share]
