Anthropic published a paper on Friday demonstrating that AI systems can reliably improve their own alignment training, a step toward the recursive...
