MIT Makes AI Training 4x Faster Without Losing Performance

Researchers found a way to make AI models smaller and faster while they're still learning, cutting training time and energy costs without sacrificing accuracy.

It's like trimming a tree as it grows instead of waiting until it's fully grown.