Sakana AI launches RSI Lab to pursue recursive self-improvement research
SakanaAILabs · x · 2026-08-04
Sakana AI announces the launch of RSI Lab, a dedicated group focused on recursive self-improvement and on redesigning the AI development process itself.
- The company argues that the next AI paradigm may come from open-ended systems that improve themselves under constraints, not just from scaling a single model.
- It frames RSI as a strategic opportunity for Japan, even if it cannot win purely on compute scale.
- The post summarizes two years of work that it says laid the foundation for the lab.
Research lineage highlighted in the article
- LLM-Squared (2024): a joint Oxford/Cambridge project where an LLM helped invent better ways to train LLMs, leading to the preference optimization algorithm DiscoPOP.
- The Darwin Gödel Machine (2025): a self-modifying agent system that evolved code changes and more than doubled the starting score on SWE-bench, gaining 30 points in absolute terms.
- ShinkaEvolve (2025): an open-source framework for efficient program evolution aimed at scientific discovery.
Related event: Sakana AI Launches RSI Lab and Joins Japan's AIRoA(2 posts)→
More from Research
- Roomer repairs 3D indoor layouts with object-grounded local edits — cn-scut · 2026-08-04
- ScrambleToolBench finds agents still brute-force tools after the map changes — declare-lab · 2026-08-04
- Hidden future trajectories make autonomous-driving VLMs reason more faithfully — Buaa1 · 2026-08-04
- WCM boosts VLA robot RL with a world-model critic and 149-task wins — OpenMOSS-Team · 2026-08-04
- StyleForge uses counterfactual reasoning to make room layouts more coherent — cn-scut · 2026-08-04
- Seedance 2.5 costs RMB 1,200 a day to test, but can synthesize industrial data — xiaohu · 2026-08-04