New
ScienceMore in Science→
Fractal Math Reveals Why AI Training Can Defy Reproducibility
Key Takeaways
- The study finds that tiny changes in AI starting conditions can produce different learned outcomes.
- The researchers link the effect to fractal boundaries between training outcomes.
- The work suggests some AI variability may be intrinsic rather than caused by hardware or data-order randomness.
DE
DT Editorial Team··via phys.org