Can AI systems drive their own progress? Recursive self-improvement research

Dongxi NLP, a Chinese NLP account on X, reports on recursive self-improvement, the idea that AI systems could improve the research that produces later AI. The post cites a coding agent that rose from 20% to 50% on a 200-task SWE-bench subset after editing its own tools, and it lays out five research routes.

The author says economic models suggest current feedback is not yet enough for self-sustaining acceleration. The post is a single report, and its figures are the author's citations rather than independently verified results.