Claude returns a confident answer that turns out to be factually wrong. This is known as:

解析

• Ais correct. "Hallucination" is the standard term for when a language model like Claude generates plausible-sounding but factually incorrect information, often stated with unwarranted confidence.It is an intrinsic characteristic of how autoregressive LLMs predict likely token sequences rather than retrieve verified facts, so it is not strictly a "bug" but a behavior to be mitigated through grounding techniques such as retrieval-augmented generation (RAG), tool/function calling, citations, and verification loops.
• Ba "context overflow" refers to exceeding the model's context-window capacity.
• Ca "temperature artefact" is not a real term — temperature governs sampling randomness/diversity, not factual accuracy.
• Da "system prompt violation" describes ignoring instructions or guardrails, not producing false facts.