Bài kiểm tra có liên kết hướng dẫn · Trung bình Cấp độ
Bài kiểm tra mô hình ngôn ngữ lớn về y tế
Kiểm tra phạm vi nhiệm vụ LLM, rủi ro ảo giác và bỏ sót, quyền riêng tư, nguyên tắc đánh giá của FDA, xác nhận và trách nhiệm giải trình của con người.
Câu hỏi 1 của 8
Why can fluent medical text still be unsafe without checking?
Xem trước mọi câu hỏi8 câu hỏi
Câu trả lời và giải thích xuất hiện khi bạn làm bài kiểm tra.
- Why can fluent medical text still be unsafe without checking?
- The model verified patient identity
- A model can produce plausible but false, incomplete, or biased content
- Fluency guarantees every fact
- Generated citations always support claims
- Which use has a different risk profile from treatment recommendation?
- Selecting a medication dose automatically
- Issuing an unreviewed diagnosis
- Changing care from a prompt
- Drafting a summary for a clinician to verify
- What should a reviewer do before signing an AI-drafted note?
- Compare it with the source record and correct errors or omissions
- Assume accuracy because it reads well
- Approve if confidence is high
- Delete the source record
- What does retrieval from approved documents guarantee?
- The model cannot omit context
- Privacy controls are unnecessary
- It can constrain sources but cannot ensure faithful generated answers
- Every statement is supported
- When may FDA clinical decision-support guidance be relevant?
- Whenever a hospital uses a word processor
- Only for open-source models
- Only for voice chatbots
- When software’s clinical function and intended use affect regulatory status and reviewability
- Which set belongs in a clinical LLM evaluation?
- Only empathy
- Omissions, contradictions, unsupported claims, privacy, and subgroup errors
- Only grammar and reading speed
- Only latency
- What privacy practice should an organization use?
- Disable access controls
- Retain data indefinitely without purpose
- Send only necessary data through an approved environment
- Paste identifiable charts into any public chatbot
- Who is responsible for a clinical decision that used generated text?
- The qualified clinician who reviews and makes the decision
- The model by itself
- The portal vendor
- The training-data authors