Bài kiểm tra có liên kết hướng dẫn · Trung bình Cấp độ

Câu đố AI sáng tạo

Đánh giá các kết quả đầu ra của AI tổng hợp theo cơ chế và cách sử dụng của chúng, tách các dữ kiện nguồn khỏi các xác nhận của mô hình và kiểm tra độ chính xác, giả tạo và rủi ro.

Đường dẫn hướng dẫn liên quanAI tạo sinh
Câu hỏi 1 của 8

Which output best illustrates generative AI as described in the guide?

Xem trước mọi câu hỏi8 câu hỏi

Câu trả lời và giải thích xuất hiện khi bạn làm bài kiểm tra.

  1. Which output best illustrates generative AI as described in the guide?
    • A newly produced summary, image, audio clip or code sample
    • A fixed list of records copied without change
    • A report showing only whether a login succeeded
    • A hand-entered table with no model output
  2. How does an autoregressive text model produce a sequence, according to the guide?
    • It retrieves a complete answer from a fixed encyclopedia
    • It predicts successive tokens using the text so far
    • It denoises random pixels into a picture
    • It converts a waveform into a class label
  3. A fluent generated summary includes a claim not found in its source. What should the reviewer do?
    • Assume fluency signals accuracy
    • Remove the source to avoid comparison
    • Separate source statements from model inference and verify the claim
    • Publish it because summaries may add context
  4. Which description matches diffusion-based image generation in the guide?
    • A system that searches the web for a matching finished picture
    • A system that copies clip art outlines
    • A system that writes one pixel row at a time
    • A system that transforms noisy representations into samples
  5. What should be checked when evaluating generated code?
    • Only whether its comments are clear
    • Whether it behaves as intended, using meaningful tests
    • Whether it was produced in one try
    • Whether the prompt contains a question mark
  6. Why should a generated citation be opened and checked?
    • Citation-shaped text can refer to a source that does not exist
    • Every model removes source details
    • A citation is always a training-data license
    • The guide says sources cannot support claims
  7. Which evaluation plan best follows the guide?
    • Use one preference score for text, code and images
    • Judge an output only by how polished it sounds
    • Match checks to the artifact, such as factual consistency for a summary and behavior tests for code
    • Skip review when the model version is new
  8. What information may help reproduce an output when that matters?
    • Only the user’s name
    • The color of the interface
    • Thời gian trong ngày
    • The model version, prompt, source material and settings