Reproduce “GDS Agent for Graph Algorithmic Reasoning” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Borun Shi, Ioannis Panagiotas
Researchers at Neo4j introduced the GDS Agent, an AI system that enables large language models to perform complex graph algorithmic reasoning on extensive graph-structured data. The agent integrates 48 graph algorithms as callable tools, demonstrating high accuracy in tool selection and parameterization, with GPT-5 achieving an 85.4% answer match on a new London Underground benchmark.
Key insights
- LLMs can effectively map natural language queries to complex graph algorithms, correctly identifying necessary tools and their parameters from a structured tool specification.
- Grounding LLM reasoning by executing validated graph algorithms on real data significantly improves the accuracy and verifiability of answers compared to ungrounded LLM-only inferences.
- Even with a robust toolset, LLMs may fail to recognize the absence of necessary data or algorithms, leading to attempts at 'pretend reasoning' with available but inappropriate tools rather than acknowledging limitations.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.