Skip to main content Skip to navigation

Can large language models provide useful feedback on research papers? A large-scale empirical analysis

Project Overview

The document examines the transformative role of generative AI, particularly large language models (LLMs) such as GPT-4, in the realm of education, with a specific focus on enhancing the peer review process for scientific research manuscripts. It addresses the escalating difficulties faced by traditional peer review systems, which are strained by the rising number of academic submissions and a lack of qualified reviewers. The findings indicate that LLM-generated feedback can effectively mirror human reviewer insights, thereby serving as a valuable complementary tool for researchers, particularly those affiliated with under-resourced institutions. This integration of AI aims to streamline the feedback process, improve the quality of manuscript evaluations, and democratize access to academic support, ultimately contributing to more equitable research opportunities in the educational landscape.

Key Applications

GPT-4 based scientific feedback generation pipeline

Context: The tool is designed for researchers submitting papers to scientific journals and conferences, specifically in fields like AI and computational biology.

Implementation: An automated pipeline was created to generate feedback on full PDFs of scientific papers by analyzing them and providing structured comments.

Outcomes: The study found that over half of the researchers found GPT-4 generated feedback helpful, with significant overlap in the points raised by LLMs and human reviewers.

Challenges: Some limitations include the LLM's tendency to focus on specific feedback aspects and lack of depth in critiques, particularly regarding methodological design.

Implementation Barriers

Technical Limitations

LLMs like GPT-4 may generate feedback that is too generic or lacks depth in specific areas, particularly in complex methodological critiques.

Proposed Solutions: Future iterations could involve fine-tuning the models for specific domains or enhancing the prompts to elicit more detailed feedback.

Access and Equity

Marginalized researchers, especially from non-elite institutions, may still face challenges in accessing timely and high-quality feedback despite the use of LLMs. Implementing LLM tools in a way that is inclusive and accessible is crucial, such as providing free access to under-resourced institutions.

Proposed Solutions: Ensure that LLM tools are made available to under-resourced institutions to promote equity in access to educational resources.

Project Team

Weixin Liang

Researcher

Yuhui Zhang

Researcher

Hancheng Cao

Researcher

Binglu Wang

Researcher

Daisy Ding

Researcher

Xinyu Yang

Researcher

Kailas Vodrahalli

Researcher

Siyu He

Researcher

Daniel Smith

Researcher

Yian Yin

Researcher

Daniel McFarland

Researcher

James Zou

Researcher

Contact Information

For information about the paper, please contact the authors.

Authors: Weixin Liang, Yuhui Zhang, Hancheng Cao, Binglu Wang, Daisy Ding, Xinyu Yang, Kailas Vodrahalli, Siyu He, Daniel Smith, Yian Yin, Daniel McFarland, James Zou

Source Publication: View Original PaperLink opens in a new window

Project Contact: Dr. Jianhua Yang

LLM Model Version: gpt-4o-mini-2024-07-18

Analysis Provider: Openai

Let us know you agree to cookies