AI is deteriorating in realtime

2026-05-20 · Source: Artificial Intelligence · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Data Science & Analytics · Depth: Advanced, medium

Summary

AI models face a significant risk of deterioration, termed "model collapse," when recursively trained on synthetic data, as highlighted by Shumailov et al. in Nature (July 2024). This issue is compounded by a projected "data drought" of high-quality human-generated content, with Gartner forecasting that 60% of training corpora will be synthetic by 2024. Villalobos et al. (ICML 2024) also discuss the limits of LLM scaling based on human data. Anecdotal evidence from data curation companies suggests widespread use of AI by "expert contributors," leading to a "bullshit in, bullshit out" scenario. While some propose autonomous agents for data orchestration and cleaning to mitigate scarcity, others critique models like MiniMax M2.7 as overhyped, emphasizing the need for data quality over mere quantity and a shift towards specialized, efficient AI agents.

Key takeaway

For AI/ML engineers developing new models, recognize the critical risk of "model collapse" from training on increasingly prevalent synthetic data. You should prioritize rigorous data provenance and quality checks, focusing on human-generated content where possible, to prevent performance degradation and increased hallucinations. Consider investing in advanced data curation techniques or specialized, efficient AI agents rather than relying solely on large, general-purpose models.

Key insights

AI models risk "model collapse" and increased hallucinations when recursively trained on synthetic data.

Principles

• Recursive training on AI-generated data degrades model performance.
• Data quality is paramount over quantity for next-generation AI datasets.
• Detecting synthetic content is challenging due to high false positive rates.

Method

Autonomous agents can orchestrate, clean, and generate data from existing sources to address data scarcity and improve quality.

In practice

• Prioritize human-generated, high-quality data for model training.
• Implement robust data provenance tracking and validation.
• Explore tightly scoped, specialized AI agents for efficiency.

Topics

• AI Model Collapse
• Synthetic Data
• Training Data Quality
• Large Language Models
• Data Scarcity
• AI Hallucinations

Best for: Research Scientist, AI Architect, AI Engineer, AI Scientist, Machine Learning Engineer, Director of AI/ML

Related on AIssential

• AI is Eating Its Own Tail And Biting The Hand That Feeds It — AI training on its own outputs creates a feedback loop leading to model collapse and degraded quality.
• Confident, Fluent, and Wrong: What Happens When Your AI Agent Trusts Bad Data — AI agents can fail dangerously by confidently presenting stale data as fact, independent of model quality.
• AI Is Eating Itself to Death — and Nobody Is Stopping It — AI models trained on synthetic data experience "model collapse," progressively losing unique information and degrading.
• Your “For You” Page Doesn’t Know You. It Predicts You. — Effective AI relies more on accurate information retrieval than on advanced generation capabilities.
• From Exams to Escape Rooms: How We Learned to Test AI — AI evaluation is a dynamic process, constantly adapting to measure increasingly complex model capabilities beyond narrow task performance.
• The Day Synthetic Data Turned Poisonous: Inside Model Collapse — Recursive training on synthetic data causes model collapse, eroding diversity and amplifying errors.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by Artificial Intelligence.