Researchers have introduced PreScience, a new dataset and benchmark designed to evaluate AI's ability to forecast future scientific advancements. The dataset comprises 98,000 recent AI research papers, along with related historical papers, creating a corpus of over 500,000 documents. PreScience supports seven forecasting tasks, including predicting future contributions, collaborator selection, and topic trend analysis. Initial experiments using various AI models, including large language models, show that generated research papers tend to be less diverse and novel compared to human-authored work. AI
IMPACT This dataset could drive research into AI's capacity for scientific prediction and innovation.
RANK_REASON The cluster contains an academic paper introducing a new dataset and benchmark. [lever_c_demoted from research: ic=1 ai=1.0]
AI-generated summary · Google Gemini · from 1 sources. How we write summaries →
