HOTSPOT (Drag and Drop is not supported)Select the answer that correctly completes the sentence.Hot Area:
Answer(s): A
Box: Natural Language Understanding (NLU) A virtual assistant that converts spoken requests into written text relies primarily on Automatic Speech Recognition (ASR) or Speech-to-Text first, while Natural Language Understanding (NLU) is used subsequently to extract meaning, intent, and entities from that generated text.How the Workload Splits Speech-to-Text (ASR): Turns audio waves into raw text words. Handles accents, background noise, and phonetic matching.Natural Language Understanding (NLU): Analyzes the text produced by the ASR engine. Figures out what the user actually wants (intent). Finds key details like dates, names, or items (entities).
https://www.assemblyai.com/blog/how-speech-to-text-ai-works-and-what-to-consider
HOTSPOT (Drag and Drop is not supported)For each of the following statements, select Yes if the statement is true. Otherwise, select No.Note: Each correct selection is worth one point.Hot Area:
Box 1: No No, a summarization model is not used to transcribe spoken audio into written text; instead, an automatic speech recognition (ASR) or speech-to-text model performs the transcription, while a summarization model is used afterward to shorten the resulting text.Box 2: No No, a standard chat completion model cannot directly generate vector embeddings for semantic tasks. While both types of models process natural language, they are distinct architectures optimized for entirely different outputs: Chat Completion Models (e.g., gpt-4o, Claude 3.5 Sonnet) are generative. They take a text prompt and predict the next sequence of words to produce a conversational text response.Embedding Models (e.g., text-embedding-3-small) are extractive. They compress text into a fixed-length numerical array (a vector) that mathematically represents the underlying meaning, or semantic context, of that text.Box 3: Yes Yes, an embedding model can generate vector representations of the provided input. It turns text, words, or images into a list of numbers. These numbers capture the meaning of the data.How Embeddings Work An embedding model maps complex data into a simple numeric space. Inputs: Words, sentences, whole documents, or images. Outputs: A dense vector of floating-point numbers. Meaning: Similar inputs get vectors that sit close to each other in that space.
https://aimultiple.com/speech-recognition https://www.debutinfotech.com/blog/understanding-the-role-of-embedding-in-models-like-chat-gpt
Box 1: No No, relying on a single method of user input does not provide an inclusive experience. True inclusivity requires supporting multiple interaction modes—such as text, voice, touch, and keyboard navigation—because a single-mode interface inevitably excludes users with specific physical, sensory, cognitive, or situational constraints.Box 2: No No, simply increasing the overall number of users does not automatically support an inclusive experience. Inclusive design requires that an AI system actively works for people of all backgrounds, abilities, and identities. High user volume can mask deep gaps where certain groups are left out or harmed.Box 3: Yes Designing an AI system for different languages supports inclusiveness by breaking down communication barriers, helping people access information easily, and making everyone feel welcome.Equal Access: Everyone gets the same chance to use the system.Clear Choices: Users can pick the language they know best.Fair Treatment: No one is left out because of how they speak or write.
https://www.mindinventory.com/blog/website-accessibility-best-practices/
Box 1: Yes Yes, limiting access to AI systems and data directly reduces the risk of unauthorized data exposure by minimizing attack surfaces and preventing accidental leaks.Key Security Benefits Fewer Entry Points: Restricting who and what can connect to your AI model lowers the chance of a security breach.Controlled Data Flow: Limiting data inputs ensures sensitive user or corporate information does not enter training pipelines or public logs.Better Compliance: Strict access rules help meet privacy laws and internal security policies.Box 2: No Sharing personal user data does not automatically improve collaboration or support responsible AI practices; in many cases, it creates significant privacy, security, and ethical risks that run counter to responsible AI principles.Box 3: Yes Yes, protecting personal data and complying with regulations are essential parts of responsible AI. They build trust, respect user privacy, and prevent illegal data use.User Trust: People want to know their private data is safe.Legal Rules: Laws punish companies that misuse data.Fairness: Good data habits stop AI from learning bad or biased habits.
https://www.linkedin.com/top-content/artificial-intelligence/ai-in-cybersecurity/data-security-issues-in-artificial-intelligence/ https://pmc.ncbi.nlm.nih.gov/articles/PMC10498316/
Box: perform consistently under expected conditions Reliability and safety is the specific Microsoft Responsible AI principle defined by the requirement that AI systems must operate reliably, safely, and consistently under expected conditions.
https://learn.microsoft.com/en-us/agents/design-guidelines/responsible-ai
Post your Comments and Discuss Microsoft AI-901 exam prep with other Community members:
Our website is free, but we have to fight against AI bots and content theft. We're sorry for the inconvenience caused by these security measures. You can access the rest of the AI-901 content, but please register or login to continue.