Mibvunzo yeAI & Kuvanzika

Mhinduro uye tsananguro zvinoonekwa paunenge uchitora bvunzo.

1. Ndechipi chirevo chinosiyanisa zvakanyanya kuchengetedza kubva kune zvakavanzika?
   • Security and privacy are identical concepts
   • Encryption alone establishes consent
   • Secure storage can still be a privacy problem if data are excessive or used unexpectedly
   • Privacy only concerns passwords
2. Mukusiyana kwekuongororwa kwekuvanzika, epsilon inopesvedzera chii?
   • The number of records in the input
   • The privacy-accuracy tradeoff, with smaller epsilon generally giving stronger privacy
   • The model’s training temperature
   • The encryption key length
3. Nhengo-inopikisa anorwisa anobvunza kana rekodhi remunhu raive mukudzidziswa. Chii chiri kuedzwa?
   • Whether a user can reset a password
   • Whether a record is duplicated in a database
   • Whether a prompt contains a name
   • Whether the model output leaks information about training-set membership
4. Sei kubvisa mazita kuchitadza kuita kuti dataset isazivikanwe?
   • Other attributes or combinations can re-identify people
   • A name is the only possible identifier
   • Data cannot be linked across tables
   • Removing names always satisfies every privacy goal
5. Ndeipi dhizaini yedata inonyanya kutevedzera gwara rekuvanzika musimboti?
   • Keep every prompt indefinitely for any future use
   • Collect only information needed for the stated task and set a retention period
   • Copy all source records into each model call
   • Treat local processing as proof of no privacy risk
6. Why should a team inspect embeddings and indexes after deleting an uploaded document?
   • Deleting a document automatically removes every copy everywhere
   • Embeddings contain no data relationships
   • Derived artifacts may still reveal information from the original data
   • An index is only a user interface
7. What should a team document about AI data processing?
   • Only the model’s public name
   • The number of support tickets
   • A generic privacy slogan
   • Which providers and internal services handle each data category
8. Why does an assistant’s unsupported claim that data were deleted not suffice?
   • The team should test deletion and account isolation through the actual application
   • A chatbot response is a deletion receipt
   • Data deletion is never relevant
   • The assistant can inspect all vendor systems