Mmastodon TechnologyAI first seen 16 h ago, last 16 h ago, peak #5
AI trained on its own output risks model collapse, critics warn
Original: When # AI is trained on its own output, or the output of other AI, it suffers model collapse. This is very unlike actual
A widely shared argument holds that artificial intelligence systems degrade — a phenomenon known as model collapse — when they are trained on their own output or on content produced by other AI systems. The point being made is that this differs sharply from human intelligence, which learns from its own actions, from peers, and from observing scenarios play out, allowing people to improve through feedback rather than deteriorate.
Why now: Growing concern about generative AI systems being trained on AI-generated content circulating online.
Evidence
- When # AI is trained on its own output, or the output of other AI, it suffers model collapse. This is very unlike actual intelligence, which routinely learns from the results of its own actions, the words and actions of its peers, and even from scenarios playing out in its own… · argv_minus_one@mastodon.sdf.org · 2
API: https://socialmediatrends-api.osmike.com/v1/trends/685174