AIF-C01 Practice Question 85: Stateless Model Conversation History
AWS Certified AI Practitioner (AIF-C01) · Test 9 · question 5 of 10
A developer builds a chat assistant that calls a foundation model on Amazon Bedrock through the API, sending only the user's latest message with each request. A user asks "What is the warranty on the X200 laptop?" and then follows up with "And does it cover the battery?" The model answers the second question as if it had never heard of the laptop. What should the developer do?
Show answer and explanation
Answer:
AWS documents that *"models don't recall prior prompts and previous requests, unless the previous interaction is included within the current prompt."* Each API call is independent, so the application must send the earlier user and assistant turns along with the new question. That gives the model the context to resolve "the battery" to the X200 (option B).
- A is wrong because temperature controls randomness in token selection. It cannot supply information the model was never given. - C is wrong because Provisioned Throughput reserves inference capacity. It does not create session memory. - D is wrong because the output token limit only caps the length of the reply. It has no effect on whether the model can see the earlier conversation.
Reference: Prompt engineering concepts