RSI is the new AGI — and it’s just as hard to pin down

The piece notes that RSI has become a three-letter byword for cataclysmic AI takeoff, with projects like Alex Karpathy's Auto-Research using agent swarms to train large language models on simple tasks. The article highlights that, like AGI, RSI remains difficult to define or predict, with no consensus on its feasibility or timeline. It positions RSI as a successor to AGI in the ongoing debate over AI's future trajectory.

It references Karpathy's project as an early example of RSI implementation, suggesting that such approaches could lead to systems that autonomously improve their own performance. The article acknowledges that the term 'RSI' carries connotations of risk, drawing on fears of exponential capability growth.

The piece places RSI in the broader context of AI development, noting that even modest advancements in self-improvement could have significant implications. It concludes that RSI, like AGI before it, is likely to remain a contested and elusive target for the foreseeable future. The article serves as a primer for readers unfamiliar with the term, outlining its origins and the questions it raises.