MENU
A team is working in Microsoft Foundry to test and compare large language model (LLM) prompt variants in a development environment.
The team requires consistent inputs to evaluate prompt variants without relying on live user traffic.
You need to create a controlled evaluation of input data.
Which action should you perform first?
Suggested Answer: A
Get Premium AI-300 Questions
as Interactive Web-Based Practice Test or PDF
Real Exam Format
Detailed Explanations
Instant Access
Get Your Practice Test
17 comments · newest first
ME
Merrilee
1 month ago
Definitely! It sets a solid foundation for testing.
LY
Lyda
1 month ago
I feel confident about starting with synthetic data.
BA
Barrett
1 month ago
Yes, and it saves time in the long run.
MY
Myrtie
2 months ago
Synthetic data can mimic real user behavior effectively.
DA
Daniela
2 months ago
True, but without data, filters are useless.
CH
Charlene
2 months ago
But what about content filters? They might be important too.
EL
Elmira
2 months ago
Agreed! It allows for controlled testing.
YO
Yolando
2 months ago
I think generating synthetic interaction data is the best first step.
DE
Denise
3 months ago
D is important too, but not the first step here.
SH
Shay
3 months ago
B and C seem more relevant for filtering out bad inputs.
KA
Karan
3 months ago
Wait, can synthetic data really mimic real user behavior?
GL
Glenna
3 months ago
Totally agree, you need that controlled input first.
JE
Jesus
3 months ago
A) Generate synthetic interaction data is the way to go!
DE
Devora
3 months ago
I feel like applying a blocklist could be useful, but it seems more like a secondary action after we have our data set up.
KI
Kimberely
4 months ago
This question reminds me of a practice question where we had to set up a controlled environment. I chose to enable observability metrics then, but I wonder if that's the right first step here.
RO
Ronald
4 months ago
I'm not entirely sure, but configuring content filters might be important too. I remember something about filtering out irrelevant data.
FR
Frederica
4 months ago
I think we should generate synthetic interaction data first since we need consistent inputs for testing.