引導式測驗 · 硬 等級

如何成為人工智慧工程師測驗

測試人工智慧工程師的學習路線圖、檢索增強生成管道的工作原理以及為什麼單獨評估檢索和生成很重要。

問題 1 的 8

根據該指南,人工智慧工程師的關注點與機器學習工程師的關注點有何不同?

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。

  1. 根據該指南,人工智慧工程師的關注點與機器學習工程師的關注點有何不同?
    • AI engineers design GPUs, while ML engineers write prompts
    • AI engineers build applications on top of existing foundation models, while ML engineers train and deploy models
    • AI engineers only do data analysis and statistics
    • There is no difference between the two roles
  2. 為什麼指南建議先學習原始 LLM API,然後再學習 LangChain 等框架?
    • Frameworks are illegal to use commercially
    • Raw APIs are always faster to write
    • Frameworks cannot call LLMs
    • It makes debugging any framework much easier later
  3. 什麼是模型上下文協定?
    • An open standard for connecting models to tools and data
    • A law regulating AI engineers
    • 一種壓縮模型權重的方法
    • A benchmark for measuring model accuracy
  4. 在 RAG 管道中,攝取過程中會發生什麼?
    • The model is retrained on each user question
    • Users rate answers from one to five stars
    • Documents are chunked, embedded and stored with metadata
    • The system deletes old documents every day
  5. 為什麼系統可以在向量搜尋的同時運行 BM25 關鍵字搜尋?
    • To catch exact keyword matches such as names and codes that embeddings may miss
    • Because vector search cannot return more than one result
    • To avoid storing any metadata
    • Because BM25 generates the final answer text
  6. What does recall at k measure in RAG evaluation?
    • How quickly the model generates each token
    • Whether the correct chunk appears among the top k retrieved results
    • How many users return to the app after k days
    • The number of words in the final answer
  7. According to the guide, where do many wrong RAG answers trace back to?
    • The user's internet connection
    • The model's temperature setting always being too low
    • The programming language used
    • Retrieval failures rather than the model itself
  8. Which misconception does the guide correct about starting in AI engineering?
    • That solid software engineering matters
    • That evaluations help catch regressions
    • That you need a PhD or advanced mathematics to begin
    • That prompt injection is a security concern