引導式測驗 · 中等 等級

ChatGPT 與法學碩士測驗

回顧下一代代幣生成、代幣、聊天機器人與模型、訓練與檢索,以及為什麼流暢的答案和引用仍然需要來源檢查。

問題 1 的 8

自回歸語言模型的常見訓練目標是什麼?

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. 自回歸語言模型的常見訓練目標是什麼?
    • 在即時驗證的資料庫中找到每個答案
    • 從前面的上下文預測可能的下一個標記
    • 將每個句子分類為真或假
    • 將每個輸入翻譯成電腦程式碼
  2. A language model can split one word into several processing units. What are those units?
    • Complete conversation turns
    • Tokens, which need not correspond to whole words
    • Final verified answers
    • Database records returned by retrieval
  3. How does a chatbot relate to an LLM?
    • A chatbot is an application that may combine a model with instructions, retrieval, tools, and an interface
    • A chatbot is always identical to one model
    • An LLM is the visual window in which a conversation appears
    • A chatbot is only a search index
  4. Which statement distinguishes training from prompting and retrieval?
    • A prompt necessarily changes model weights
    • Retrieval permanently modifies the model
    • A tool call is the same as a training epoch
    • Training adjusts parameters, while a prompt supplies current context and retrieval provides selected sources
  5. Does supplying sources eliminate hallucinations?
    • Yes, because every retrieved passage is always followed exactly
    • Yes, as long as the source is public
    • No; a model can misread, ignore, or miscite supplied evidence
    • No; retrieval prevents all useful answers
  6. What is a reliable way to check a factual claim with a citation?
    • Trust it if the title appears plausible
    • Open the cited source and confirm it supports the claim
    • Ask the same model whether the citation is real
    • Check only whether the URL loads
  7. A note says a workshop starts at 10:00 and registration closes Friday. What time does the workshop end?
    • The note does not state an ending time
    • 11:00
    • Noon
    • The time the chatbot was opened
  8. What does the original Transformer paper establish about every commercial chatbot?
    • That all products use an identical implementation
    • That every chatbot uses the same training data
    • That all models have the same context limit
    • It describes an architecture, but does not establish every product’s implementation details