CCA-F Question 53

Community Discussion

Written by other learners. Comments can disagree with the marked answer — check the official documentation when in doubt.

• Chose C1

  Speed: Shrinks the prompt size by condensing older turns, avoiding the slowdown from 50k tokens. Quality: Keeps the most recent 5–6 turns verbatim, ensuring precision in immediate context while summaries preserve the essence of earlier exchanges. User experience: Users don’t need to re‑state past information all the time apart from exact numerical , date or status values (unlike sliding windows), and the assistant still feels like it “remembers” the conversation. Balance: Unlike retrieval, which risks missing context if semantic search fails, summarization guarantees continuity while staying efficient.

• Chose D1

  Given that, a sliding window that keeps the system prompt plus the last 8-10 turns (comfortably covering the 94% case with margin) and gracefully asks for restatement in the rare 6% case is the most proportionate solution. It directly attacks both symptoms: response speed (token count per request drops dramatically, so generation is faster) and coherence (the model isn't straining to attend over 50k tokens of increasingly diluted context, which is a known contributor to degraded output quality in long contexts).

• Chose C2

  From the user's perspective C looks better