Copyright (c) 2026 MindMesh Academy. All rights reserved. This content is proprietary and may not be reproduced or distributed without permission.

8.2. Common Exam Traps and Distinctions

The AI-102 exam includes several common misconceptions that trip up candidates. Review these traps carefully—understanding why the wrong answers are wrong is as important as knowing the correct answer.

Service Selection Traps

TrapCorrect Understanding
OCR is NLPOCR is Computer Vision (input is image)
Document Intelligence = OCRDocument Intelligence extracts structured data, not just text
DALL-E can analyze imagesDALL-E generates images; use GPT-4o Vision for analysis
Question answering handles dynamic dataOnly works with static information

Configuration Traps

TrapCorrect Understanding
Lower temperature = more creativeLower temperature = more deterministic
Content being blocked? Change categoriesAdjust severity levels, not categories
Substitution errors in STTAdd custom vocabulary (names, products)
BLEU 30 is high quality40-59 is high quality
Special characters OK in synonymsSpecial characters (#, @, %) are not allowed

Security Traps

TrapCorrect Understanding
Containers work offlineStill need billing endpoint
API keys are secure enoughManaged identity is most secure
All networks is fine for productionUse Private endpoint for Azure-only access

API/Implementation Traps

TrapCorrect Understanding
DALL-E needs size in bodyOnly prompt is required in body
Translator needs one headerNeeds both Ocp-Apim-Subscription-Key AND Ocp-Apim-Subscription-Region
Password-protected PDF size issuePassword protection always fails regardless of size
Alvin Varughese
Written byAlvin Varughese
Founder15 professional certifications