Generative AI Flashcards

(42 cards)

1
Q
A
How well did you know this?
1
Not at all
2
3
4
5
Perfectly
2
Q

What is Generative AI?

A

Generative AI is a subset of Artificial Intelligence where an AI model creates original content in response to a natural language prompt, producing new outputs such as text, images, and code.

How well did you know this?
1
Not at all
2
3
4
5
Perfectly
3
Q

How do Large Language Models (LLMs) work?

A

LLMs work by recognizing patterns and making predictions based on vast amounts of text data. When given a prompt, an LLM predicts the next most likely word, adds it to the sequence, and repeats the process to generate coherent text.

How well did you know this?
1
Not at all
2
3
4
5
Perfectly
4
Q

What is the difference between LLMs and Small Language Models (SLMs)?

A

LLMs have billions or trillions of parameters for comprehensive language generation but can impact performance and are difficult to deploy locally. SLMs are trained on smaller, focused datasets, making them effective for specific topics and easier to fine-tune and deploy locally.

How well did you know this?
1
Not at all
2
3
4
5
Perfectly
5
Q

What are the two main components of a transformer model?

A
  1. Encoder - reads and understands input text, picking up meanings of words and their context. 2. Decoder - generates new text based on the encoder’s understanding, creating coherent sentences.
How well did you know this?
1
Not at all
2
3
4
5
Perfectly
6
Q

What is tokenization in transformer models?

A

Tokenization is the process of breaking down input text into smaller pieces called ‘tokens’ (words, parts of words, punctuation, emojis), each assigned a unique numeric identifier to allow computers to process and understand language.

How well did you know this?
1
Not at all
2
3
4
5
Perfectly
7
Q

What are embeddings in AI models?

A

Embeddings are special numeric codes (vectors) that each token is converted into after tokenization. These vectors capture the semantic meaning of words, where words with similar meanings have similar codes in multidimensional space.

How well did you know this?
1
Not at all
2
3
4
5
Perfectly
8
Q

What is positional encoding and why is it important?

A

Positional encoding is a technique that ensures the model retains the order of words in a sentence, as word order significantly impacts meaning in natural language.

How well did you know this?
1
Not at all
2
3
4
5
Perfectly
9
Q

What are attention layers in transformer models?

A

Attention layers help the model determine the importance of each word or token within a sentence in relation to other words. ‘Self-attention’ considers how other tokens influence a token’s meaning, while ‘multi-head attention’ applies multiple perspectives for richer understanding.

How well did you know this?
1
Not at all
2
3
4
5
Perfectly
10
Q

What is natural language generation in Generative AI?

A

Natural language generation is creating human-like text responses, such as cover letters, healthy breakfast ideas, or email drafts, based on natural language prompts.

How well did you know this?
1
Not at all
2
3
4
5
Perfectly
11
Q

How does AI image generation work?

A

AI image generation interprets natural language requests to generate appropriate images, like creating an elephant eating a plant-based burger or a florist business logo based on text descriptions.

How well did you know this?
1
Not at all
2
3
4
5
Perfectly
12
Q

What can AI code generation do for developers?

A

AI code generation assists software developers by writing code snippets, explaining code, adding documentation, refactoring, optimizing code, and generating test cases.

How well did you know this?
1
Not at all
2
3
4
5
Perfectly
13
Q

What is Azure OpenAI Service?

A

Azure OpenAI Service is Microsoft’s cloud solution for deploying, customizing, and hosting large language models from OpenAI, combining cutting-edge models with Azure’s security and scalability.

How well did you know this?
1
Not at all
2
3
4
5
Perfectly
14
Q

What are GPT-4 and GPT-3.5 models used for?

A

GPT-4 and GPT-3.5 models generate text and programming code from natural language prompts. GPT-3.5 Turbo is specifically optimized for conversations.

How well did you know this?
1
Not at all
2
3
4
5
Perfectly
15
Q

What are DALL-E models designed for?

A

DALL-E models are designed to create and edit images from text descriptions.

How well did you know this?
1
Not at all
2
3
4
5
Perfectly
16
Q

What is the purpose of embedding models in Azure OpenAI?

A

Embedding models convert text into numerical sequences for analysis and comparison of text similarity.

How well did you know this?
1
Not at all
2
3
4
5
Perfectly
17
Q

What does the Whisper model do?

A

Whisper transcribes and translates speech to text.

18
Q

What is Azure OpenAI Studio?

A

Azure OpenAI Studio is a web-based environment for deploying, testing, and managing large language models (LLMs).

19
Q

What are Copilots in AI?

A

Copilots are generative AI assistants integrated into applications, often as chat interfaces, to provide contextualized support for common tasks and boost productivity and creativity.

20
Q

What does Microsoft Copilot for Microsoft 365 do?

A

Microsoft Copilot for Microsoft 365 assists with creating documents, spreadsheets, presentations, and managing emails.

21
Q

How does GitHub Copilot help developers?

A

GitHub Copilot aids software developers with real-time code suggestions, documentation, and testing.

22
Q

What is the function of Microsoft Bing Search Engine Copilot?

A

Microsoft Bing Search Engine Copilot generates natural language answers to search queries.

23
Q

Who does Copilot for Azure assist?

A

Copilot for Azure assists infrastructure administrators in the Azure portal.

24
Q

What does Microsoft Copilot for Security help with?

A

Microsoft Copilot for Security helps security professionals assess, mitigate, and respond to threats.

25
What is Microsoft Fabric Copilot used for?
Microsoft Fabric Copilot helps analysts generate code for data analysis, manipulation, and visualization in Spark Notebooks and Power BI.
26
What options do organizations have for implementing Copilots?
Organizations can use off-the-shelf copilots, extend existing ones to support custom business processes, or build entirely custom copilots.
27
What is prompt engineering?
Prompt engineering is the process of refining prompts (instructions) given to an AI application to generate higher-quality and more targeted responses.
28
What is the purpose of defining a system message in prompt engineering?
A system message sets the context, expectations, and constraints for the model's behavior.
29
What are the three types of shot learning in prompt engineering?
1. Zero-shot learning (no examples provided), 2. One-shot learning (one example provided), 3. Few-shot learning (several examples provided) to guide the AI.
30
What is grounding in prompt engineering?
Grounding is providing specific, relevant context within a prompt (e.g., including an email text to be summarized) to help the AI produce more accurate and related responses.
31
What are the four stages of Microsoft's responsible generative AI development process?
1. Identify potential harms relevant to the solution, 2. Measure the presence of harms in outputs, 3. Mitigate harms at multiple solution layers, 4. Operate with deployment and operational readiness plans.
32
How does Microsoft's responsible AI approach relate to industry standards?
Microsoft's responsible generative AI framework aligns with the NIST AI risk management framework.
33
What features does Azure Content Safety offer?
Azure Content Safety offers prompt shields (scanning for user input attacks), groundedness detection, and protected material detection.
34
What is Azure AI Studio?
Azure AI Studio is a unified development environment that brings together capabilities from Azure Machine Learning, Azure OpenAI Service, and other Azure AI services for building comprehensive AI solutions.
35
What can developers create in Azure AI Studio?
Developers can create and manage AI projects and hubs for centralized setup and collaboration.
36
What does the model catalog in Azure AI Studio include?
The model catalog includes models from Microsoft, OpenAI, Hugging Face, Meta, and other providers that can be deployed and tested.
37
What is prompt flow in Azure AI Studio?
Prompt flow is a feature designed to streamline the entire lifecycle of LLM-powered applications, offering visual graphs, prompt variants, and evaluation tools.
38
What is retrieval augmented generation (RAG)?
RAG (Retrieval Augmented Generation) is the integration with custom data sources for augmenting prompts, allowing AI models to access and use specific organizational data.
39
What monitoring capabilities does Azure AI Studio provide?
Azure AI Studio allows developers to monitor and evaluate AI models and prompt flows to ensure performance and quality.
40
What can be built using prompt flow in Azure AI Studio?
Developers can build generative AI applications and copilots using prompt flow within Azure AI Studio.
41
What deployment capabilities does Azure AI Studio offer?
Azure AI Studio allows developers to deploy and test LLMs from the model catalog, providing a comprehensive platform for model management.
42
What makes Azure AI Studio comprehensive for AI development?
Azure AI Studio brings together multiple Azure AI services in one environment, enabling end-to-end development from model selection to deployment and monitoring.