"The last invention that man need ever make": The risk of self-improving AI

Researchers at Anthropic and OpenAI say AI training has grown more automated, nearing the recursive self-improvement threshold. An OpenAI employee told The Information the company has largely automated training new experimental models, with AI systems running experiments as directed by humans. Anthropic shared that AI leads about 26% of its R&D work and collaborates on 90% of work, raising safety concerns.