An AI team is organizing a table of when each model selection criterion should be prioritized. In which situation is it MOST appropriate to prioritize task fit (accuracy for the use case)?
Choosing the situation where prioritizing task fit is appropriate.
When you want to keep the bill for a large volume of requests as low as possible.
When you want to keep the bill low, the criterion to prioritize is cost (price per token).
This situation maps to the cost criterion rather than task fit, so it is incorrect.
When you want to confirm whether the model can meet the quality required for that use case.
Correct. In model selection, the fundamental step is to first confirm task fit, that is, whether the model can meet the quality required for that use case.
When you want to prioritize how quickly a response begins to return above all else.
When you prioritize response speed, the criterion to prioritize is latency.
This is not a situation that maps to task fit, so it is incorrect.
When you want to handle inquiries in multiple languages with a single model.
When handling multiple languages, what you check is multilingual support.
This is not a situation that maps to task fit, so it is incorrect.
In model selection, the fundamental step is to confirm 'whether the model can meet the quality required for the use case (task fit/accuracy).' Cost, latency, and context window are then balanced once the required quality is met.