5.4 Describe How Token Usage and Context‑window Management Affect Prototyping Cost, Latency, and Output Quality AI-TECHNICAL-PRACTITIONER Practice Questions (Page 2)

Questions 6–10

1. 6application · easy

   A developer is comparing two prompts for an LLM API. Prompt A is 200 tokens, and Prompt B is 400 tokens. Both produce the same quality output. The developer wants to reduce cost and latency. What should they do?

   • AUse Prompt A because fewer input tokens reduce cost and latency.
   • BUse Prompt A but increase the max output tokens.
   • CUse Prompt B because longer prompts improve model reasoning.
   • DUse Prompt B but lower the temperature.
   Reveal Solution Discussion Select an answer first
2. 7expert · hard

   A real-time translation app sends the entire user conversation (about 3,000 tokens) plus a 500-token instruction prompt to the LLM. The app needs to respond in under 2 seconds. The current latency is 3 seconds. Which change is most likely to bring latency under 2 seconds without sacrificing translation quality?

   • AIncrease the max tokens for the response to allow longer translations.
   • BUse a model with a larger context window.
   • CReduce the instruction prompt to 100 tokens and trim the conversation to the most recent 1,000 tokens.
   • DSet the temperature to 0.0 for more deterministic output.
   Reveal Solution Discussion Select an answer first
4. 8application · medium

   A developer is using an LLM to answer questions about a long research paper. The paper is 15,000 tokens, and the model's context window is 16,000 tokens. The developer sends the entire paper plus a question. The model's answer misses key details from the middle of the paper. What is the most likely cause and best fix?

   • AThe model's temperature is too high; lower it to 0.0.
   • BThe output is being truncated; increase max tokens to 2,000.
   • CThe paper is too long; truncate the paper to the first 5,000 tokens.
   • DThe prompt is too close to the context limit, causing the model to lose focus; use a retrieval approach to feed only relevant sections.
   Reveal Solution Discussion Select an answer first
5. 9expert · hard

   A developer is fine-tuning a chatbot that must remember user preferences across a long session. The context window is 4,000 tokens. After 30 minutes, the conversation history is 3,800 tokens. The user asks a question that requires remembering a preference mentioned in the first message. The model fails to recall it. What is the most effective way to fix this while keeping the session within the context window?

   • ATruncate the conversation to the last 10 messages.
   • BSwitch to a model with a larger context window.
   • CIncrease the max tokens for the response to 2,000.
   • DUse a running summary of the conversation that includes key preferences, and replace the full history with the summary plus recent messages.
   Reveal Solution Discussion Select an answer first
6. 10foundation · easy

   How does a larger token count in a request affect latency?

   • AIt increases latency because the model must process more tokens.
   • BIt has no effect on latency; latency is only affected by network speed.
   • CIt only affects latency if the request exceeds the context window.
   • DIt decreases latency because the model can better understand the request.
   Reveal Solution Discussion Select an answer first