The first experimental evidence of recursive self-improvement
AIDE² researchers successfully demonstrated recursive self-improvement, where an outer loop optimizes an inner-loop agent, resulting in search algorithms that outperform hand-tuned systems.
Deep dive
- Inner Loop: Standard autoresearch agent optimizing code.
- Outer Loop: Meta-optimizer adjusting the harness configuration based on average performance across benchmarks.
- Key Discoveries: Novel search policies, 16x prompt compression, and automated layered defenses against reward hacking.
- Generalization: The discovered agent outperformed a two-year hand-tuned baseline on held-out benchmarks, including physics-based weather modeling.
Decoder
- Autoresearch: The practice of using LLMs to iteratively write, test, and improve code or research processes.
- Recursive self-improvement (RSI): The theoretical concept where an intelligent system can improve its own design or intelligence, potentially leading to rapid capability jumps.
Original article
The first experimental evidence of recursive self-improvement (RSI).
Autoresearching the autoresearch agent for eight days.
The result beats the harness we hand-tuned for two years, on held-out benchmarks.
Our RSI system AIDE² has two autoresearch loops.
An inner loop, just like a normal autoresearch agent, optimizing code against an eval.
An outer loop, optimizing the inner-loop agent's harness code against the inner loop's average score across different benchmarks.
After 100 iterations, the outer loop discovered seven improvements over the baseline.
Including a new search policy, a memory system that compresses prompt by 16x, and a layered defense against reward hacking.
We test the discovered agents on held-out benchmarks the outer loop never saw.
They generalize. They beat the agent we hand-tuned for two years, on all three.
Two sit inside its training task families. The farthest sits outside, improving a physics-based weather model.
We also see an emergent phenomenon where the outer loop pushes the inner-loop agent's reward hacking rate lower, with a combination of prompting and rule-based checks.
This was benchmarked on OOD GPU kernel engineering tasks that suffered from reward hacking.
On our RSI ladder, AIDE² is Level 1.
Its self-improvement efficiency went beyond manual R&D with general AI tools, on held-out benchmarks.
We also tested Level 2, whether the improved inner agent makes a better outer loop. Results are mixed, and we do not claim ignition.
More in the blog post:
- a breakdown of the discovered algorithms
- the rejected ideas AIDE² tried, covering a surprising share of the search literature
- the dead code it shipped
Very proud of the team, Dhruv Srikanth, Yuxiang Wu, Dex Hunt, and Bingchen Zhao, for shipping such an ambitious project spanning nearly a year with relatively few resources.
Also, a huge thank you to everyone who provided feedback on the draft, including Jean Kaddour, Minqi Jiang, Morgan McGuire, odysseus0z, Ross Taylor, Ofir Press and many others!
Fresh dev news, AI-digested daily. Never miss a trend at devoured.fyi