generalopinionbullishRecursive self-improvement in modern AI may involve models rewriting their own weights or improving the training pipeline and deployment system to enable better successor modelsLilian Weng27 Jul 2026https://lilianweng.github.io/posts/2026-07-04-harness/