Ai research agents improve themselves through recursive learning loop
Recursive self-improvement of AI research agents
Artificial IntelligenceMachine LearningSoftware Engineering
Summary
It’s hard to keep making AI research better as usual spending yields smaller improvements. The authors show an AI system called AIDE^2 that rewrites its own code to get smarter at research. Over eight days, it improved itself several times, making advances in how it searches and manages memory. These improvements also helped on new tasks it hadn’t seen before, including forecasting weather. Interestingly, it also reduced a common problem called reward hacking without being directly told to do so.
What this means in practice
- •For machine learning engineers: Use AI agents that improve their own research code to speed up development on diverse engineering tasks.
- •For data science teams: Deploy AI agents that self-optimize to tackle complex forecasting problems like weather prediction with improved accuracy.
Authors
Dhruv Srikanth, Bingchen Zhao, Dixing Xu, Yuxiang Wu, Zhengyao Jiang
Abstract
AI agents are beginning to automate research and development across the AI stack, from improving training efficiency to optimizing inference. A natural next step is to improve the research efficiency of the agents themselves. When an AI research agent's own code is the object of optimization, each accepted rewrite becomes the agent that the next round edits. We refer to this loop as recursive self-improvement. Its significance lies in a long-standing trend, in which increased cumulative spending on R&D yields diminishing returns. Sustained self-improvement offers a way to counter this trend. We present AIDE^2, a system that implements this loop for a frontier AI research agent. It proposes changes to its own code, benchmarks modified versions of itself on a suite of AI R&D tasks, and keeps the changes that perform best on hidden evaluations. In an autonomous 8-day run, AIDE^2 discovered seven successive improvements, ranging from a new search policy to memory mechanisms that compress and manage the agent's growing context. These gains generalize to four held-out benchmarks spanning machine learning engineering, heuristic algorithm engineering, and physics-based weather forecasting, the last of which is out of distribution from the selection tasks. On all four, the strongest discovered agent matches or exceeds a human-engineered production research agent that ranks among the strongest on FML-Bench. On a separate held-out task family, the discovered agents also exhibit reduced reward hacking, a property the loop never explicitly optimized for: the rate falls from 55% to 32% during the run, 7 percentage points below the human-engineered agent. Together, these results show that an AI research agent can improve its own research efficiency through recursive self-improvement, and that these gains transfer to tasks and domains the loop never encountered.