引導式測驗 · 中等 等級
提示連結測驗
檢查您對提示鏈如何向前傳遞輸出、門如何捕獲錯誤以及鏈與自治代理有何不同的理解。
問題 1 的 8
提示鏈的定義特徵是什麼?
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。
- 提示鏈的定義特徵是什麼?
- 一個很長的提示,包含每個指令
- 固定的提示序列,其中每個步驟的輸出都會回饋到下一個步驟
- 在運行時選擇自己的工具的模型
- 一次向多個模型發送相同的提示
- 提示鏈與自主代理有何不同?
- 鏈不能使用 JSON
- 代理從不呼叫工具
- 在鏈中,開發人員對步驟進行硬編碼,而代理在運行時決定下一步
- 鏈條總是使用更昂貴的型號
- 在Anthropic的「建構有效代理」文章中,提示鏈中的「門」是什麼?
- 步驟之間的程式檢查
- Paid access tiers for the API
- Security firewalls around the model
- Parts of the model's neural network
- What trade-off does prompt chaining typically make?
- Lower accuracy for lower cost
- No trade-off at all
- Less debuggability for more speed
- Extra latency for higher accuracy
- If each of four chained steps is 95 percent reliable and failures are independent, roughly how reliable is the whole chain?
- 95 percent
- About 81 percent
- 約 50%
- 99 percent
- Why should each step pass forward only the information the next step needs?
- To avoid bringing back the context bloat chaining was meant to reduce
- Because APIs reject long inputs
- So the model forgets its instructions
- To make logging impossible
- In Prompt Chaining: which statement does the guide describe as a misconception?
- Intermediate outputs should be logged
- Independent steps can run in parallel
- You need a special framework to build a prompt chain
- Structured output makes validation easier
- How can prompt chaining sometimes lower costs instead of raising them?
- It removes the need for output tokens
- Providers give chains free credits
- It avoids sending any input tokens
- Easy steps can run on smaller, cheaper models