Tarea 5.2 - Prompt Sensitivity and Stability in LLMs
Requisitos de finalización
Abrió: miércoles, 5 de marzo de 2026, 00:00
Cierre: jueves, 19 de marzo de 2026, 14:00
Instructions:
-
Select 5 factual or reasoning questions.
-
Create 3 paraphrased versions of each question.
-
Query the same model with all variants.
Measure: -
Answer consistency
-
Accuracy differences
-
Confidence variation
Questions to answer:
-
Does phrasing affect correctness?
-
Is reasoning invariant to surface form?
-
Which prompts produce instability?
Deliverable:
Short report + comparison table.