ASI-Evolve: AI Accelerates AI
Summary: arXiv:2603.29640v1 Announce Type: new
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.
- 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.
Key Discoveries of ASI-Evolve
- Neural Architecture Design: ASI-Evolve discovered 105 state-of-the-art (SOTA) linear attention architectures. The best-discovered model surpassed DeltaNet by +0.97 points, representing nearly three times the gain of recent human-designed improvements.
- Pretraining Data Curation: The evolved pipeline improves average benchmark performance by +3.96 points, with gains exceeding 18 points on the Massive Multitask Language Understanding (MMLU) benchmark.
- 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.
Broader Implications
ASI-Evolve is not limited to improving AI technologies; it also provides initial evidence that this AI-for-AI paradigm can transfer beyond the AI stack. Experiments conducted in mathematics and biomedicine have shown promising results, suggesting that the principles underlying ASI-Evolve can be applied in various scientific domains.
Together, these results suggest that ASI-Evolve represents a promising step toward enabling AI to accelerate AI across the foundational stages of development. The framework offers early evidence for the feasibility of closed-loop AI research, which could fundamentally change the way we approach AI development in the future.
Conclusion
The emergence of ASI-Evolve signifies a pivotal moment in AI research. By harnessing the potential of AI to enhance its own development, researchers can potentially overcome some of the significant challenges that have historically hindered progress in the field. As more results emerge from this innovative framework, the future of AI research may be transformed, ushering in a new era of accelerated discovery and innovation.