Open-source AI tool beats LLMs in literature reviews – and gets citations right
a month ago
- #AI
- #Open Source
- #Scientific Research
- Nature.com recommends using an up-to-date browser for the best experience.
- OpenScholar is an AI model that reviews scientific literature better than some major LLMs and matches human accuracy in citations.
- OpenScholar links information directly to a database of 45 million open-access articles to prevent hallucinations.
- Unlike many commercial tools, OpenScholar is open-source, allowing free use and deployment by researchers.
- AI firms like OpenAI have adopted similar methods to improve their LLMs' accuracy.
- OpenScholar is cost-effective compared to commercial alternatives like GPT-5 with deep research.
- Limitations of OpenScholar include not always retrieving the most relevant papers and being constrained by its database scope.
- LLMs often struggle with citations due to their training on non-scientific sources and generating text based on probable but incorrect associations.
- An analysis found 51 papers at NeurIPS 2025 contained non-existent or inaccurate citations.