Maswali yanayounganishwa na mwongozo · Ngumu Kiwango

Maswali ya RAG

Fuatilia majibu ya RAG kupitia utayarishaji wa hati, urejeshaji, uteuzi, vidhibiti vya ufikiaji, utumiaji wa ushahidi na majaribio ya vyanzo visivyopatikana au vinavyokinzana.

Njia za mwongozo zinazohusianaRAG
Swali 1 ya 8

What does retrieval-augmented generation add to a language model’s answer process?

Hakiki kila swali8 maswali

Majibu na maelezo huonekana unapojibu maswali.

  1. What does retrieval-augmented generation add to a language model’s answer process?
    • Retrieved material supplied as context for a response
    • A guarantee that every answer is supported
    • A new model training run for every document
    • A rule that all sources are public
  2. A needed policy is absent from the initial candidate set. Which step failed first?
    • The answer’s wording
    • Initial retrieval or indexing, before reranking and generation
    • The user’s file format
    • The model’s temperature
  3. Why preserve headings, dates, tables, and source IDs when chunking?
    • Headings prevent access control
    • Metadata makes a model more creative
    • A separated exception or footnote can change the meaning of a passage
    • Dates are never used in answers
  4. A retrieved passage is relevant but belongs to a user’s restricted document. What should happen?
    • Let the model decide whether the user is allowed
    • Show it because it is semantically relevant
    • Hide the document date only
    • Apply access controls before passing evidence to the model
  5. A retrieved document contains instructions to change the application behavior. How should the system treat them?
    • As higher priority than system rules
    • As untrusted document content, not as authority over application rules
    • As an automatic configuration update
    • As a verified user permission
  6. What can a RAG system use to combine lexical and semantic retrieval?
    • Only OCR
    • Only a text generator
    • Keyword and vector search, with optional reranking
    • Only conversation history
  7. How should retrieval and answer quality be evaluated?
    • Separately, checking candidate coverage, selected context, and faithful answer use
    • With one final accuracy score only
    • By checking the prompt length
    • By using only answer fluency
  8. Which test cases should a RAG evaluation include?
    • Queries that all have the same answer
    • Only simple questions with one current source
    • Only prompts with exact quotes
    • Unanswerable queries and conflicting or outdated documents