Reproduce “Diagnostic Foundation for Evaluating LLMs' Research Integrity as Co-Scientists” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Yash Tripathi, Silu Sharma, Sai Sidhanth Manoharan Jayanthi, Shivank Garg, Lin Li
IntegrityBench evaluates whether language models distinguish research misconduct from legitimate practices under pressure. It pairs 18 misconduct behaviors with ethical controls and tests classification, hypothetical reasoning, and artifact-grounded decisions. Misconduct-task scores averaged 73.5 under implicit versus 68.8 under explicit pressure; ethical-control scores were 61.4 versus 65.2. These benchmark scores do not measure real-world misconduct.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.