LLMs' impact on science: Booming publications, stagnating quality
a day ago
- #Research Productivity
- #Peer Review
- #AI in Science
- High-profile cases of scientific papers retracted due to AI-generated content raise concerns about peer review quality.
- Researchers from Berkeley and Cornell analyzed pre-publication papers to detect AI influence.
- They found increased paper production and improved language quality with AI use, but a drop in publication rates.
- The study involved scanning 1.2 million documents from arXiv, 675,000 from SSRN, and 220,000 from bioRxiv.
- A model was trained to distinguish human-generated text from AI-generated text using pre-ChatGPT abstracts.
- The research identified a transition point when authors first used LLMs, noting a significant increase in scientific output post-adoption.