Abstract Multi-agent LLM systems are deployed on the assumption that agents checking one another reduce hallucination. Two studies of the same linear topology contradict it in opposite directions,...
<p><strong>Tác giả:</strong> Shaoxuan Zhang</p>
<p>Abstract Multi-agent LLM systems are deployed on the assumption that agents checking one another reduce hallucination. Two studies of the same linear topology contradict it in opposite directions, and both are right: one measures prevalence, the other detectability, and a claim can become rarer and less visible at once. Whether a false claim grows is set by two structural variables: how correlated the agents' errors are, and whether any channel supplies evidence independent of them. Topology...</p>
<p>📄 DOI: <a href="https://doi.org/https://doi.org/10.21203/rs.3.rs-11536507/v1" target="_blank" rel="noopener">https://doi.org/10.21203/rs.3.rs-11536507/v1</a></p>
<p>📖 <a href="https://doi.org/https://doi.org/10.21203/rs.3.rs-11536507/v1" target="_blank" rel="noopener">Xem bài báo đầy đủ</a></p>
Bài viết được biên soạn lại bởi LIFVietnam. Nguồn tham khảo: OpenAlex
