Study Finds Language Models Show Conformity Bias in RAG When Majority Endorses Poisoned Claims
A new arXiv paper examines whether increasing the number of retrieved documents that endorse a misleading claim makes language models more likely to adopt it in retrieval-augmented generation.
Read the full story at arXiv cs.CL ↗
Timeline · 1 report
- 2026-11-02 05:00 · arXiv cs.CL
Language Models Show Conformity Bias in RAG: When the Majority Endorses a Poisoned Claim