Why LLMs invent answers instead of saying they don't know
LLMs generate plausible-sounding but incorrect answers because they lack a true understanding of knowledge boundaries. Unlike humans who can say "I don't know," LLMs are designed to always produce a response, leading to confabulation rather than simple hallucination. This distinction matters for how we build trust in AI systems.
背景メモ
- 大規模言語モデル（LLM）は、答えを知らないときに「わからない」と言うことができず、自信満々に間違った情報を生成する「ハルシネーション（幻覚）」と呼ばれる現象を起こす。