#machine-unlearning

The new open-weight model reduces flagged censorship responses from 89.8% to 2.8% while maintaining general benchmarks within ~1 point of the original.

Reddit·2026.10.08 19:00

Removing low-influence data first in machine learning models can cut unlearning costs by up to 50%.

Apple ML·1·2026.07.17 09:00