Reproduce “Fluent but Fallible: Benchmarking Hallucination and Factual Reliability in Large Language Models” 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
Osama Shabih, Dayam Nadeem, Shahab Saquib Sohail, Mohammad Nadeem
The authors compare ten language models on 70 academic prompts across four difficulty categories, evaluating both incorrect claims and omitted information using reliability measures. Mean accuracy was lowest for Advanced prompts (0.608) and highest for Tricky prompts (0.955); the authors suggest Tricky’s true/false format may explain this pattern, which is limited to one February 2025 run per prompt.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.