Anthropic shows AI can train other AI models far cheaper than humans

This could slash the huge compute and labeling bills we see when fine‑tuning models. If the trick works reliably, smaller teams might get alignment‑safe models without burning cash. Still needs solid benchmarks to trust the results.