An AI practitioner is comparing foundation models on Amazon Bedrock for an internal FAQ chatbot (Japanese only, text only, no additional model tuning needed). The speed at which a response starts coming back drives the experience, and because a large volume of questions is processed continuously, suppressing the bill is also necessary. Which TWO selection criteria directly correspond to these requirements? (Choose TWO.)

1 / 1
Select all that apply
CorrectA, C

Explanation

Question Overview

Select two selection criteria that directly correspond to the scenario's requirements.

Requirements to satisfy
  • 1「speed at which a response starts coming back」The immediacy requirement = look at latency
  • 2「suppressing the bill is also necessary」Token consumption accumulates = look at inference cost
Per-option explanation
ACorrect

Latency

This is correct. Latency is the criterion that represents the speed at which a response starts coming back. The chat experience is determined by the wait time until the first response, so it directly corresponds to this question's requirement that 'speed drives the experience.'

BIncorrect

Multilingual support

Multilingual support is a valid selection criterion that represents the breadth of languages handled, and is emphasized when you want to respond to users in multiple languages.

However, this question explicitly states 'Japanese only,' and most models satisfy the requirement, so it is not a deciding factor for narrowing candidates, making it incorrect.

CCorrect

Inference cost

This is correct. Inference cost is a criterion related to the bill, such as the per-token price. Processing a large volume of questions continuously accumulates token consumption, so it directly corresponds to this question's requirement of 'suppressing the bill.'

DIncorrect

Multimodal support

Multimodal support (modality) is a valid criterion that represents the types of input and output handled, and is checked with top priority when handling images or audio.

However, this question explicitly states 'text only,' and image/audio support is not part of the requirement, so it is incorrect.

EIncorrect

Customizability

Customizability is a valid criterion that represents how easily the model can be tuned for your own use, such as through fine-tuning, and is emphasized when you want to build in proprietary behavior or style.

However, this question explicitly states 'no additional model tuning needed,' so it is not a deciding factor, making it incorrect.

Key Takeaway

Model selection criteria—'cost / latency / modality / multilingual support / model size / customizability'—are all valid, but the point of this format is to select only those that directly correspond to the scenario's requirements. Here, 'speed = latency' and 'bill = inference cost' apply, and the premises of Japanese only, text only, and no tuning are the conditions that rule out the remaining criteria.

Related Links