Anthropic unveils ai-driven method for training AI, offering cost and speed advantages

Anthropic has introduced an automated system that uses AI to train other AI models, significantly reducing costs and increasing efficiency. The system, led by researcher Chen Yuehan, improves alignment performance across 10 benchmarks without compromising model capabilities. It operates by reviewing literature, proposing training methods, and iteratively refining results over 30 minutes. The automated researcher outperformed human experts, achieving results in just 6 hours. The system costs about $4 per hour, compared to $150 for human researchers. While promising, the method relies on accurate benchmarks and continuously updated literature. The research highlights the potential for recursive self-improvement in AI, raising questions about the future role of human researchers.