Reproduce “CACTUS: Chemistry Agent Connecting Tool-Usage to Science” 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
Andrew D. McNaughton, Gautham Ramalaxmi, Agustin Kruel, Carter R. Knutson, Rohith A. Varikoti, Neeraj Kumar
CACTUS (Chemistry Agent Connecting Tool-Usage to Science) integrates open-source large language models with specialized cheminformatics tools, enabling them to accurately address complex chemistry problems. The system significantly outperforms standalone LLMs on various chemistry tasks and functions effectively across diverse hardware.
Key insights
- Tool augmentation significantly enhances the capabilities of large language models (LLMs) in specialized scientific domains like chemistry, enabling them to tackle complex problems that neither component could solve independently.
- Domain-specific prompting strategies improve the performance of tool-augmented LLMs, particularly for qualitative chemistry tasks requiring greater contextual understanding.
- Smaller, open-source LLMs such as Gemma-7b and Mistral-7b, when augmented with specialized tools, can achieve high accuracy comparable to larger models, making advanced chemical AI more accessible across diverse hardware configurations.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.