ASI-Evolve: AI Accelerates AI

2026-03-31

AI Summary

This work addresses the challenge that AI systems struggle to autonomously drive real-world research cycles characterized by high costs, long durations, and weak supervision. To overcome this, the authors propose a unified agent framework that enables automated, synergistic exploration of the three core components—data, architecture, and algorithms—through a closed-loop “learn-design-experiment-analyze” pipeline. The framework integrates evolutionary algorithms, a cognitive foundation infused with human priors, an analysis module driven by experimental feedback, and a multi-agent collaboration mechanism, substantially enhancing exploration efficiency and knowledge reuse. Empirically, the approach discovers 105 state-of-the-art linear attention architectures, achieves a performance gain of over 18 points on MMLU, and yields a novel reinforcement learning algorithm that significantly outperforms existing methods across multiple benchmarks, with preliminary evidence of cross-domain generalization.

Abstract

Can AI accelerate the development of AI itself? While recent agentic systems have shown strong performance on well-scoped tasks with rapid feedback, it remains unclear whether they can tackle the costly, long-horizon, and weakly supervised research loops that drive real AI progress. We present ASI-Evolve, an agentic framework for AI-for-AI research that closes this loop through a learn-design-experiment-analyze cycle. ASI-Evolve augments standard evolutionary agents with two key components: a cognition base that injects accumulated human priors into each round of exploration, and a dedicated analyzer that distills complex experimental outcomes into reusable insights for future iterations. To our knowledge, ASI-Evolve is the first unified framework to demonstrate AI-driven discovery across three central components of AI development: data, architectures, and learning algorithms. In neural architecture design, it discovered 105 SOTA linear attention architectures, with the best discovered model surpassing DeltaNet by +0.97 points, nearly 3x the gain of recent human-designed improvements. In pretraining data curation, the evolved pipeline improves average benchmark performance by +3.96 points, with gains exceeding 18 points on MMLU. In reinforcement learning algorithm design, discovered algorithms outperform GRPO by up to +12.5 points on AMC32, +11.67 points on AIME24, and +5.04 points on OlympiadBench. We further provide initial evidence that this AI-for-AI paradigm can transfer beyond the AI stack through experiments in mathematics and biomedicine. Together, these results suggest that ASI-Evolve represents a promising step toward enabling AI to accelerate AI across the foundational stages of development, offering early evidence for the feasibility of closed-loop AI research.

Problem

• AI-for-AI
• long-horizon research
• weakly supervised learning
• neural architecture search
• automated discovery

Innovation

• agentic AI
• AI-for-AI
• evolutionary framework
• cognition base
• closed-loop research