Maswali yanayounganishwa na mwongozo · Ngumu Kiwango

Maswali ya Utawala wa Data ya AI

Hufanyia majaribio ruhusa mahususi za matumizi, ukoo, mali inayotokana, kufuta, umiliki wa taratibu, masahihisho na ukaguzi unaoendelea wa ufikiaji.

Njia za mwongozo zinazohusianaUtawala wa Takwimu wa AI
Swali 1 ya 8

A dataset may be used for training and evaluation. What must be checked?

Hakiki kila swali8 maswali

Majibu na maelezo huonekana unapojibu maswali.

  1. A dataset may be used for training and evaluation. What must be checked?
    • Whether its file opens in the training pipeline
    • Whether its permissions cover each intended use separately
    • Whether both uses require the same model
    • Whether the dataset has a large number of rows
  2. Why does data governance track embeddings and cached responses?
    • They contain no source-related information
    • They are always public
    • They cannot be deleted
    • Derived assets may retain information or dependencies from source data
  3. A deletion procedure removes the uploaded file but not its index copies. What remains unresolved?
    • The model’s user interface
    • The source file’s format
    • Deletion across relevant derived stores and copies
    • The data catalog title
  4. Who should approve changes to data schema and validation rules?
    • An assigned owner responsible for quality and change control
    • Any user who uploads a file
    • Mfano wa AI yenyewe
    • The storage vendor only
  5. What does lineage provide?
    • Proof that data use is permitted
    • A guarantee of data quality
    • A substitute for retention rules
    • A way to trace where data and derived outputs came from
  6. A dataset was collected for retrieval but is now proposed for model training. What is needed?
    • Assume access for one use implies every use
    • Review whether its permissions and purpose cover training
    • Delete the data immediately
    • Change the dataset label only
  7. What should a correction procedure specify?
    • Who handles it, how it propagates, and how completion is verified
    • Only where the data catalog is stored
    • The model’s average token count
    • Whether an error appears in a presentation
  8. Why revisit data access and retention when integrations change?
    • Integrations never affect data flow
    • Retention applies only before deployment
    • New services can change who sees the data and how long it persists
    • A new connector automatically narrows access