Bài kiểm tra có liên kết hướng dẫn · Cứng Cấp độ
Tỷ lệ ảo giác trong bài kiểm tra về công cụ AI hợp pháp
Kiểm tra cách các nghiên cứu của Stanford xác định và đo lường ảo giác AI hợp pháp, câu trả lời sai là gì và tại sao việc tạo tăng cường truy xuất vẫn thất bại.
Câu hỏi 1 của 8
How did the Stanford "Hallucination-Free?" study define a hallucination?
Xem trước mọi câu hỏi8 câu hỏi
Câu trả lời và giải thích xuất hiện khi bạn làm bài kiểm tra.
- How did the Stanford "Hallucination-Free?" study define a hallucination?
- Any answer longer than the source document
- An answer that is either incorrect or misgrounded
- Only a citation to a case that does not exist
- Any answer the tool took more than a minute to produce
- In the Stanford study's terms, what is a 'misgrounded' answer?
- An answer that refuses to respond
- An answer written in the wrong language
- An answer with no citations at all
- An answer that cites a real source that does not support its claim
- Which tool performed best in the 2024 Stanford test but still hallucinated on more than 17 percent of queries?
- Lexis+ AI
- Westlaw AI-Assisted Research
- Ask Practical Law AI
- GPT-4 used as a comparison
- What hallucination range did "Large Legal Fictions" find for general-purpose chatbots on verifiable federal court questions?
- 1 to 3 percent
- 17 to 33 percent
- 69 to 88 percent
- 95 to 100 percent
- Why can a retrieval step find the wrong authority even when the database contains the right one?
- Legal databases are encrypted and cannot be searched
- Retrieval only works for statutes, not cases
- Retrieval ranks documents alphabetically
- Retrieval matches similar wording or meaning, which is not the same as legal relevance
- What tendency did the research find when questions contained a false premise?
- Models tended to accept the premise and answer as if it were true
- Models always corrected the premise
- Models refused every such question
- Models replied only with the text of the relevant statute
- Which mitigation specifically helps flag a cited case that was later reversed or overruled?
- Raising the model's temperature setting
- Making answers shorter
- Translating the query into another language
- Connecting a citator such as KeyCite or Shepard's
- What marketing claim did the Stanford study's title directly question?
- That legal AI tools were cheaper than paralegals
- That some tools were 'hallucination-free'
- That legal AI could pass the bar exam
- That AI tools were approved by courts