The concept of recursive self-improvement (RSI) has gained traction in AI research, with multiple startups and researchers pursuing the goal of creating AI systems that can continuously upgrade themselves. Richard Socher, a well-known AI researcher, launched the Recursive Superintelligence project with RSI as its primary objective. 'Our main focus is to build truly recursive, self-improving superintelligence at scale,' Socher stated, emphasizing the automation of the entire research process. Other prominent researchers, including Alex Karpathy, are also exploring RSI through projects like Auto-Research, which uses agent swarms to train large language models on simple tasks. Karpathy has been transparent about the project, sharing milestones and code through a public GitHub repository. While the work so far has focused on minor improvements to a GPT-2 scale model, it has sparked interest among other researchers. Adaption, founded by Sara Hooker, recently launched AutoScientist to automate frontier training, aiming to simplify the process of training large-scale models. Doris Xin of Disarray highlighted the challenge of reliability, noting that while infinite compute and time could theoretically enable RSI, current systems still require human oversight. Meanwhile, Google CEO Sundar Pichai acknowledged that while progress is being made, the industry is not yet at the RSI threshold. Anthropic's Claude Code has shown significant self-improvement capabilities, with nearly 100% of its code written by the tool. However, it still faces challenges in self-direction and understanding organizational priorities. The AI industry remains divided on the timeline for achieving RSI, with some experts suggesting it may be closer than others. *Source: [techcrunch](https://techcrunch.com/2026/05/28/rsi-is-the-new-agi-and-its-just-as-hard-to-pin-down/)*