DiscoveryBench: Towards Data-Driven Discovery with Large Language Models
What this work does
Constructs discovery questions and reference hypotheses from published data-analysis workflows, with a separate synthetic dataset-generation route.
Connection to discovery
A fixed-data benchmark for hypothesis formulation, dataset interpretation and multistep analysis.
Evidence
The body defines contexts, variables and relationships, and evaluates their recovery rather than treating every plausible explanation as correct.
Scope and limitations
Published-reference matching narrows what counts as discovery, and reference-guided oracle feedback is a special condition. Existing scientific data do not constitute new physical experiment execution.
THE RESEARCH CONVERSATION
Discussion & reactions
Share a question, an insight, or a replication experience. Sign in with GitHub to join the conversation.