Researchers show AI systems can train other AI systems faster — speeding up the development cycle from months to weeks

A research team published ASI-Evolve, a method where one AI system generates training data and guidance to teach another AI system, compressing what normally takes months of human-labeled data collection into a fraction of the time.

Instead of humans hand-labeling thousands of examples, the first AI generates synthetic training examples and evaluates the second AI's performance, then auto-corrects — similar to a teacher writing their own practice problems and grading student work instantly.

For AI labs and companies building AI products, this means new models can be trained and deployed 3–5× faster, letting startups and established tech firms iterate on features (better chatbots, reasoning engines, code assistants) without waiting for massive labeling teams, fundamentally changing how competitive speed-to-market becomes in AI development.