On the caveats of AI autophagy

X **ng, F Shi, J Huang, Y Wu, Y Nan, S Zhang… - Nature Machine …, 2025 - nature.com
Generative artificial intelligence (AI) technologies and large models are producing realistic
outputs across various domains, such as images, text, speech and music. Creating these …

Universality of the Pathway in Avoiding Model Collapse

A Dey, D Donoho - arxiv preprint arxiv:2410.22812, 2024 - arxiv.org
Researchers in empirical machine learning recently spotlighted their fears of so-called
Model Collapse. They imagined a discard workflow, where an initial generative model is …

Enhancing Virtual Try-On with Synthetic Pairs and Error-Aware Noise Scheduling

N Li, KJ Shih, BA Plummer - arxiv preprint arxiv:2501.04666, 2025 - arxiv.org
Given an isolated garment image in a canonical product view and a separate image of a
person, the virtual try-on task aims to generate a new image of the person wearing the target …

Machine-generated text detection prevents language model collapse

G Drayson, V Lampos - arxiv preprint arxiv:2502.15654, 2025 - arxiv.org
As Large Language Models (LLMs) become increasingly prevalent, their generated outputs
are proliferating across the web, risking a future where machine-generated content dilutes …

A Theoretical Perspective: When and How Self-consuming Training Loops Generalize

S Fu, Y Wang, Y Chen, X Tian, D Tao - The Thirteenth International … - openreview.net
High-quality data is essential for training large generative models, yet the vast reservoir of
real data available online has become nearly depleted. Consequently, models increasingly …