Neural Networks Now Write Their Own Training Data, Redefining AI Research
Researchers at three major labs have independently confirmed that large language models can generate synthetic training datasets that improve their own performance, raising questions about recursive self-improvement and the future trajectory of artificial intelligence development.