출처가 붙는 연구 답변
이걸 믿을 필요는 없습니다. 모든 진술이 출처 논문을 보여줍니다.
Language models produce fluent text whether or not it is true, and fabricated references are the most consequential form that takes in academic use. They look correct: plausible author names, a real journal, a volume and page range in the right format. Several documented cases exist of fabricated citations reaching court filings and published papers, because nobody opened them. The only defence that scales is an answer that shows its sources, so checking is a click rather than a search. The point is not that you should trust a sourced answer more — it is that you no longer have to decide whether to trust it.
Why do language models fabricate citations?
They generate text that fits the pattern of a citation rather than retrieving a record. A plausible reference and a real one look identical.
How do I verify a citation quickly?
Open it. If the source is linked, verification is a click; if it is only described, you are doing a search to confirm something you were told.
Should I cite the tool in my paper?
Cite the papers. Many journals also ask you to disclose AI assistance separately, which is a different requirement from citation.