Chinese researchers publish roadmap toward recursive self-improvement, 'The Last AI Built by Humans'
kimmonismus · x · 2026-09-14
As US frontier labs debate slowing AI development, researchers from Shanghai Jiao Tong, Theseus Labs, Tsinghua, ByteDance and others have published 'The Last AI Built by Humans', a roadmap toward recursive self-improvement.
- Maps five stages, from executing improvements to changing the process that produces future improvements
- Authors stress it's a research roadmap with preliminary evidence, not a working system anyone can download
- The ambition: AI that gets better at making itself better
The author's takeaway: the coordination problem is now in plain view — US labs can slow down, but cannot assume everyone else will follow.
Related event: Chinese Team Releases 75-Page Roadmap for Recursive Self-Improving AI(14 posts)→
More from AGI Musings
- Why AI risk coordination may be easier than climate: shorter timescales, earlier action — moultano · 2026-09-15
- Neil Chowdhury joins Thinking Machines to work on safety and alignment — ChowdhuryNeil · 2026-09-15
- DeepSeek fans compare Dario to Hitler as AI safety-vs-accel camp feud turns nasty — zephyr_z9 · 2026-09-15
- Mathematician on AI learning: a correct route still won't teach you the map — AlexKontorovich · 2026-09-15
- Alpha School Now Runs 9 Themed AI Schools, From Esports to Wilderness Survival — thesaraharminta · 2026-09-15
- "Products barely work, yet everyone's racing to add guardrails": data scientist pushes back on AI safety hype — kareem_carr · 2026-09-15