NVIDIABased on real interviews
Choose between faster and stronger summaries
Determine where a slower model's better offline summaries would be worth the extra wait.
Unlock with Full AccessThe challenge
An analyst tool can switch to a model that produces stronger offline summaries but takes about three times as long. Its users range from people scanning a report during live work to researchers preparing something for later review.
The quality results have not been separated by task or user group. Work out what the team needs to learn before changing the default, using the available product events and options for streaming, routing and experimentation.
What you'll practise
Quality evidence
Ask what the comparisons establish for the specific tasks and users that matter.
Response experience
Connect the cost of waiting to the job the analyst is trying to complete.
Explain how the answers would change the decision or scope.