Reproduce “NONSENSE: Non-Semantic Strings Expose Semanticity Gaps in Dense Retrieval” 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
Ankit Gupta
NONSENSE audits whether dense retrievers can rank meaningless strings above relevant evidence. It optimizes fixed-length character strings by repeatedly changing one character to increase query similarity while rejecting strings containing query terms. On SciFact with E5, strings beat the strongest judged relevant passage on 0.977 of queries, but the result varies across model–dataset pairs and does not establish downstream harm.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.