Harness Engineering for Self-Improvement
Lilian Weng Blog · Lilian Weng · 2026-07-04
Lilian Weng's blog post traces recursive self-improvement from I.J. Good's 1965 ultraintelligent machine concept to modern frontier AI labs, where models are beginning to improve the training pipelines and systems that produce their successor versions.
Appears in
Extraction
Topics: recursive-self-improvementai-capabilitiesfrontier-aiai-research-acceleration
Claims
- The concept of recursive self-improvement dates to I.J. Good's 1965 definition of an ultraintelligent machine that can design better versions of itself.
- Modern recursive self-improvement can take the form of models improving training pipelines and deployment systems rather than directly rewriting their own weights.
- The speed of AI research development has demonstrably accelerated at frontier labs including Anthropic and OpenAI.
Key quotes
an AI uses its current intelligence to improve the cognitive machinery that produces its intelligence
The speed of research development in AI has been shown to drastically accelerated in frontier labs (Anthropic; OpenAI).