Reproduce “Data-driven Discovery with Large Generative Models” 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

Bodhisattwa Prasad Majumder, Harshit Surana, Dhruv Agarwal, Sanchaita Hazra, Ashish Sabharwal, Peter Clark

This position paper explores the use of Large Generative Models (LGMs) for end-to-end data-driven scientific discovery, proposing a hybrid system that combines LGM capabilities with robust external tools and active human feedback. Their proof-of-concept, DATAVOYAGER, demonstrated the potential for automated hypothesis generation and verification from existing datasets while also highlighting the necessity of human oversight and tool integration to mitigate LGM limitations.

Key insights

  • Large Generative Models (LGMs) exhibit significant potential for comprehensive data understanding, hypothesis generation, and multi-step planning in scientific discovery.
  • LGMs alone are insufficient for reliable end-to-end data-driven discovery due to limitations such as hallucinations, constrained 'System-2' reasoning, and challenges with human alignment.
  • A hybrid approach integrating LGMs with robust, 'fail-proof' analytical tools and active user moderation through feedback mechanisms is crucial for accurate and reliable automated discovery.

Summary by alphaXiv.

Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.