The first experimental evidence of recursive self-improvement (3 minute read)

# Thread by @zhengyaojiang on Thread Reader App Source: [https://threadreaderapp.com/thread/2077079778793042425.html](https://threadreaderapp.com/thread/2077079778793042425.html) [](https://threadreaderapp.com/user/zhengyaojiang) The first experimental evidence of recursive self\-improvement \(RSI\)\. Autoresearching the autoresearch agent for eight days\. The result beats the harness we hand\-tuned for two years, on held\-out benchmarks: 🧵\(1/7\)  Our RSI system AIDE² has two autoresearch loops\. An inner loop, just like a normal autoresearch agent, optimizing code against an eval\. An outer loop, optimizing the inner\-loop agent's harness code against the inner loop's average score across different benchmarks\. \(2/7\)[](https://pbs.twimg.com/media/HNNEKFkbMAADV_d.jpg) After 100 iterations, the outer loop discovered seven improvements over the baseline\. Including a new search policy, a memory system that compresses prompt by 16x, and a layered defense against reward hacking\. \(3/7\)[](https://pbs.twimg.com/media/HNNETVsbcAA6J6s.jpg) We test the discovered agents on held\-out benchmarks the outer loop never saw\. They generalize\. They beat the agent we hand\-tuned for two years, on all three\. Two sit inside its training task families\. The farthest sits outside, improving a physics\-based weather model\. \(4/7\)[](https://pbs.twimg.com/media/HNNEZCAa4AERHPz.jpg) We also see an emergent phenomenon where the outer loop pushes the inner\-loop agent's reward hacking rate lower, with a combination of prompting and rule\-based checks\. This was benchmarked on OOD GPU kernel engineering tasks that suffered from reward hacking\. \(5/7\)[](https://pbs.twimg.com/media/HNNEm2IbYAAtR_j.jpg) On our RSI ladder, AIDE² is Level 1\. Its self\-improvement efficiency went beyond manual R&D with general AI tools, on held\-out benchmarks\. We also tested Level 2, whether the improved inner agent makes a better outer loop\. Results are mixed, and we do not claim ignition\. \(6/7\)[](https://pbs.twimg.com/media/HNNEuSVb0AA15MD.jpg) More in the blog post: \- a breakdown of the discovered algorithms \- the rejected ideas AIDE² tried, covering a surprising share of the search literature \- the dead code it shipped \(7/7\)[weco\.ai/blog/first\-evi…](https://www.weco.ai/blog/first-evidence-of-recursive-self-improvement) Very proud of the team, @DhruvSrikanth, @yuxiangwu\_, @dexhunt3r, and @BingchenZhao, for shipping such an ambitious project spanning nearly a year with relatively few resources\. Also, a huge thank you to everyone who provided feedback on the draft, including @jeankaddour, @MinqiJiang, @morgymcg, @odysseus0z, @rosstaylor90, @OfirPress and many others\! • • • Missing some Tweet in this thread? You can try to[force a refresh](https://threadreaderapp.com/thread/2077079778793042425.html#)