Advanced · Research

Model Collapse

Visual diagram · (in preparation) · Math · (in preparation) · Worked example · 3 difficulty levels.

TL;DR. A hypothesized phenomenon where AI models continually trained on data generated by other AI models degrade in quality over generations.

Technical Definition

A hypothesized phenomenon where AI models continually trained on data generated by other AI models degrade in quality over generations.

How it works

Model collapse is a concern for the long-term sustainability of AI training. If training data increasingly consists of synthetic content generated by previous AI models, the diversity and quality of the data could diminish. This could lead to models that forget nuance, overfit to artificial patterns, and ultimately lose generalizability and accuracy, hindering future AI development.

Related Concepts

• Synthetic Data — Artificially generated data mimicking real-world properties, used for training augmentation or privacy protection.
• Data Drift — When the statistical distribution of inputs to a deployed model changes over time.