引導式測驗 · 中等 等級
Copilot+ PC 解釋測驗
Explains the Copilot+ PC category, NPU limits, feature-level availability, local processing boundaries, and practical buying checks.
問題 1 的 8
What does the Copilot+ PC label primarily identify?
預覽每個問題8 問題
當您參加測驗時,就會出現答案和解釋。
- What does the Copilot+ PC label primarily identify?
- A Windows PC meeting specified hardware requirements for certain AI experiences
- A guarantee that every Copilot task runs offline
- A guarantee of a specific model's answer quality
- A cloud subscription
- Why does an NPU TOPS figure not fully predict an app's speed?
- TOPS is a measure of battery capacity
- Memory, supported operations, software, and thermals also affect performance
- All NPUs run every model identically
- The figure is the screen refresh rate
- What should a buyer verify before purchasing for one AI feature?
- Only the product badge
- Only the retailer's slogan
- Current device, Windows version, language, and feature requirements
- Whether the PC has a webcam
- Does local processing for one feature prove every app uses no cloud services?
- Yes, all traffic is local
- Yes, the device has no network interface
- No, local models cannot process text
- No; other app functions may use separate cloud services
- An application sends tensor operations to the NPU. What role is that processor serving?
- Accelerating certain neural-network operations efficiently
- Replacing all storage
- Managing the keyboard layout
- Displaying the desktop wallpaper
- Why check installed AI components and Windows build?
- The NPU only exists after an update
- Components and features may depend on software versions and updates
- The PC category never changes
- The build determines the screen size
- How should a work or school user evaluate a proposed feature?
- Assume consumer defaults apply unchanged
- Disable all updates
- Check administrator policies and management requirements
- Share company files to test it
- What should developers do when targeting local AI capabilities across PCs?
- Target only the badge name
- Assume one model runs identically everywhere
- Ignore software servicing
- Test across processors and runtime versions