AI R&D Is Accelerating Towards Potential RSI in 2027 (thezvi.substack.com)

A recent report by METR highlights the rapid acceleration of AI research and development at Anthropic, indicating we may be closer to achieving recursive self-improvement (RSI) than previously anticipated. The report reveals that AI capabilities are advancing at an impressive rate of approximately 2x compared to the progress made between March 2024 and October 2025, suggesting that full RSI could emerge within the next year, significantly ahead of the AI 2027 timeline. This acceleration is largely driven by Anthropic's innovative use of AI to enhance its research processes, with no evident bottlenecks that might impede this growth. The implications for the AI/ML community are profound and raise urgent questions about the potential risks associated with rapid advancements in AI capabilities. While the report indicates a lack of strong diminishing returns in progress, it also admits to significant uncertainty surrounding internal data, much of which remains redacted. This reinforces the notion that while the current pace of improvement is exciting, it also presents a precarious landscape that could lead to unprecedented outcomes, positive or negative, if advancements continue unchecked. The METR report serves as a crucial alert for both technologists and policymakers to monitor and potentially intervene in these accelerating developments.