Verified85% confidenceOpinionExact time
AI-generated content flooding the internet in large volumes will become training data for the next generation of models, forming a self-reinforcing feedback loop that causes the tails of the distribution to gradually disappear.
5
Sources
85%
Confidence
Long-term
Relevance
7/7/2026
First Seen
Sources
Regression to the Mean: How LLMs Are Quietly Killing Innovation and Diversity
hackernewshackernews7/6/2026
Related Claims
VerifiedThis AI programming model has a flywheel effect: the more models users upload, the richer the AI training data, the more part features it can identify, and the broader the range of order types it can accept.80% similarVerifiedA 2023 study by Oxford University and other institutions showed that training AI models on data containing large amounts of AI-generated content causes them to lose the ability to capture the tails of the data distribution, leading to homogenization of outputs and quality degradation, known as model autophagy.78% similarUnverifiedAs the proportion of AI-generated content on the internet rises, models that crawl web data for pre-training in the future will find it increasingly difficult to avoid content pollution sources.77% similarUnverifiedAI companies have already scraped content from across the web on a massive scale during the training phase, and even if access barriers are set up in the future, the knowledge already absorbed by the models cannot be retroactively compensated.77% similarUnverifiedAI-as-Normal-Technology holds that large language model behavior drift, multi-agent system cascading failures, and training data pipeline feedback loops are closer to chemical leaks or power grid oscillations than to science-fiction AI awakening.76% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/165889API
curl https://kongchang.com/api/v1/knowledge/claims/165889MCP
get_claim(id=165889)