TagsAlphaResearch: Accelerating New Algorithm Discovery with Language Models (Nov 2025)
- Link: http://arxiv.org/abs/2511.08522v1
- Date: November 2025
Abstract
This paper introduces AlphaResearch, an autonomous research agent leveraging large language models (LLMs) to discover new algorithms for open-ended problems. It proposes a novel dual research environment combining execution-based verification with a simulated real-world peer-review system. AlphaResearch iteratively generates ideas, verifies them, and optimizes proposals. The paper also presents AlphaResearchComp, a new benchmark of eight open-ended algorithmic problems, showing AlphaResearch achieving best-of-known performance on one problem (‘packing circles’) and a 2/8 win rate against human researchers, demonstrating LLMs’ potential in accelerating algorithm discovery.
Key Topics:
- Autonomous Research Agents
- Algorithm Discovery
- Large Language Models (LLMs)
- Dual Research Environment
- Program-based Verification
- Peer Review Simulation
- Open-ended Algorithmic Problems
- AlphaResearchComp Benchmark
- Packing Circles Problem