33 Researchers Publish 75-Page Roadmap Toward Genuine Recursive Self-Improvement

33 researchers including Bowen Zhou and Zhiyuan Liu published a 75-page arXiv paper, "The Last AI Built by Humans: Toward Genuine Recursive Self-Improvement," systematically addressing recursive self-improvement (RSI) and proposing an implementation path. It attracted significant attention as a roadmap-level report.

Confirmed

The paper has 33 co-authors from Shanghai Jiao Tong University, Tsinghua, ByteDance, Shanghai AI Lab and TheseusLabs; per Jiqizhixin, ModelBest and Xiaohongshu also participated, and AI-Cambrian's retelling mentions SenseTime among others
Based on a survey of 72 companies/teams including OpenAI, Anthropic and Google DeepMind, TheseusLabs frames it as a five-level RSI framework, while the paper presents an autonomy-centered five-stage roadmap
Core question: AI can already execute improvement work; whether it can genuinely improve itself and bypass the human-labor bottleneck is the key issue
Motivating background: continually expanding parameter counts and training costs of frontier models
Steven Byrnes shared the report; Plinz warned about anti-AI sentiment in the US in the same thread; Chinese tech media and overseas commentators widely relayed it

Why it matters

The paper directly tackles the frontier debate on whether AI can achieve genuine self-improvement
teortaxesTex noted it refutes claims in parts of the EA community that Chinese researchers don't believe in superintelligence
kimmonismus and others read it as a US-China divergence: while US frontier labs debate slowing down, a Chinese-led team has published a roadmap toward RSI, with geopolitical undertones highlighted by Plinz's promotion amid rising anti-AI sentiment in the US

2026-09-13 ~ 2026-09-15 · 14 related posts

Primary sources

Researchers map a roadmap toward genuine recursive self-improvement in AI — irinarish · 2026-09-14 08:42 (UTC+08:00)
33-author paper maps the road to recursive self-improvement, reviewing 72 AI teams — huybery · 2026-09-13 04:56 (UTC+08:00)
33-author arXiv paper lays out a roadmap toward genuine recursive self-improvement — chaumian · 2026-09-13 04:56 (UTC+08:00)

[source] 33-author arXiv paper lays out a roadmap toward genuine recursive self-improvement — chaumian · 2026-09-13
[source] 33-author paper maps the road to recursive self-improvement, reviewing 72 AI teams — huybery · 2026-09-13
Paper on recursive self-improvement: AI that gets better at getting better — teortaxesTex · 2026-09-13
SJTU, Tsinghua, ByteDance paper lays out 5-stage roadmap to recursive self-improvement — Dr_Singularity · 2026-09-14
New Report Outlines Autonomy-Centered Roadmap for Genuine Recursive Self-Improvement in AI — Dan_Jeffries1 · 2026-09-14
TheseusLabs maps recursive self-improvement with a 5-level RSI framework from 72 industry practices — 机器之心 · 2026-09-14
75-Page Survey Maps Five Levels of Recursive Self-Improvement in AI — AI寒武纪 · 2026-09-14
Chinese researchers publish roadmap toward recursive self-improvement, 'The Last AI Built by Humans' — kimmonismus · 2026-09-14
Source paper link for 'The Last AI Built by Humans' recursive self-improvement roadmap — kimmonismus · 2026-09-14
As US Labs Debate Slowing Down, Chinese Researchers Publish Roadmap to Recursive Self-Improvement — kimmonismus · 2026-09-15

4 near-duplicate retellings: Dr_Singularity · irinarish · debashis_dutta · debashis_dutta