"LLMs as Research Tools: A Large Scale Survey of Researchers' Usage and Perceptions"

Survey reveals how 816 researchers actually use LLMs, exposing surprising demographic adoption patterns.

This survey paper studies how researchers across different demographics and fields use LLMs in their research workflows, and their perceptions about benefits, risks, and ethical concerns.

https://arxiv.org/abs/2411.05025

Results 📊:

→ Information Seeking (49%) and Editing (45%) are most frequent LLM use cases

→ Data Analysis (31%) and Generation (27%) are least used

→ Grammar/phrasing fixes most common (50% of users)

→ Significant demographic variations in adoption: Non-White researchers 2.1x more likely to use LLMs frequently

-----

Key Insights from this Paper 💡:

→ 81% of researchers have incorporated LLMs into their research workflow

→ Non-White, junior, and non-native English speaking researchers show higher LLM usage and perceived benefits

→ Women, non-binary, and senior researchers express greater ethical concerns

→ Computer science researchers show higher comfort with LLM usage disclosure

→ Researchers strongly prefer open-source/non-profit LLM tools over commercial ones