Reproduce “PaperTrail: A Claim-Evidence Interface for Grounding Provenance in LLM-based Scholarly Q&A” 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
Anna Martin-Boyle, Cara A.C. Leckey, Martha C. Brown, Harmanpreet Kaur
PaperTrail introduces a claim-evidence interface for LLM-based scholarly question-answering systems, explicitly grounding AI-generated claims to their source documents to enhance provenance. The system significantly reduced users' subjective trust in LLM outputs by highlighting supported, unsupported, and omitted claims, but revealed a "trust-behavior gap" where this reduced trust did not translate into increased editing, often due to system latency and usability challenges.
Key insights
- Argument-grounded provenance, by explicitly mapping claims to evidence, significantly reduces users' subjective trust in LLM outputs, encouraging increased caution and skepticism.
- A 'trust-behavior gap' was identified, where reduced trust in LLM outputs due to detailed provenance did not necessarily lead to greater behavioral reliance (more editing), largely constrained by time pressure and usability issues.
- The backend architecture offers a framework for objectively evaluating LLM trustworthiness, serving as a benchmark for faithfulness and completeness, particularly in detecting critical information omissions.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.