Practice Test Generative AI Foundations

The Generative AI Foundations practice test contains 110 questions and covers the following objectives:

Generative AI Methods and Methodologies -  23 questions

1.1 Define Generative AI.

• Compare and contrast Generative AI with predictive AI, discriminative AI, analytical AI, statistical AI.
• Compare and contrast Generative AI with search engines.
• Fundamental understanding of diffusion model, transformer model, generative adversarial networks (GANs), variational autoencoders (VAEs).

1.2 Explain the basic processes Generative AI uses to produce an output.

• Understand that each model is trained differently.
  • Text models include: OpenAI GPTx, Google Gemini, Anthropic Claude, Meta LLaMA
  • Image models include: DALL-E, Adobe Firefly
• Large language models (LLMs) need a large amount of training data to perform effectively.
• LLMs are trained on such a huge dataset that there will be opinions and points of view.
• Image models are trained on text-image pairs that are manually tagged.
• Training a model consumes a large amount of energy and requires powerful GPUs.

1.3 Recognize the input and output types used in a Generative AI scenario.

• You can use multiple inputs to get an output.
  • Inputs include: text, audio, video, images
  • Output types include: generative text, generative video, generative image, generative audio
• Different tools allow different types of input to generate an output.

1.4 Recognize that Generative AI models can be customized to perform individualized tasks.

• Self-contained app that does a task for you.
  • Examples: Custom GPT, Google Gems, Microsoft Copilots

1.5 Select an appropriate tool to perform a specific task

• Tools: Microsoft Copilot, Google Gemini, MetaGPT, Adobe Express, Canva, Open AI ChatGPT, Claude, Microsoft Azure AI Studio, Stable Diffusion.
• Considerations for selecting a tool: purpose and functionality, ease of use, cost, updates and support, data privacy, security, quality, customizability, parameters available for output control.

1.6 Describe the limitations of Generative AI.

• Output is not reliable.
• Output could include bias, misinformation, and hallucinations.
• Needs processing power and access to the data (usually internet).
• Conversations are used for training unless you enable privacy settings.
• No universal standards on how it should be used.
• Limitations with consistency (two clocks, each show a different time).
• Rapid changes might make previous work obsolete.

Basic Prompt Engineering - 33 questions

2.1 Identify appropriate prompts to elicit textual information.

• Content gathering.
• Summarization.
• Content creation and ideation.

2.2 Identify appropriate prompts to transform content.

• Reformatting content to meet a requirement.
• Editing and proofreading documents.
• Providing a visualization of content.
• Transforming content into a different media type.
• Translating content to a different language.
• Personalizing and adapt content to facilitate learning and comprehension.

2.3 Identify appropriate prompts to elicit image creation and transformation.

• Producing an image for a specific purpose.
• Exploring artistic ideas.
• Transforming an image.
• Describing the content of images.

2.4 Identify appropriate prompts to elicit video creation and transformation.

• Adding motion to images.
• Interpolating between images.
• Colorizing a black and white movie.
• Generating video from a prompt.
• Generating an avatar that reads a script.
• Adding and removing objects in a video.
• Automated subtitling.

Prompt Refinement - 33 questions

3.1 Given an initial prompt and its output, evaluate how the prompt can be improved to elicit more targeted output.

• Content
  • Creating a prompt at the right level of specificity.
  • Creating prompts that are clear and not abbreviated.
  • Not making the assumption that the AI will “know” what you’re talking about.
• Style
  • Including information about the style and tone of the output.
  • Including a style guide.
• Persona
  • Giving the AI a persona or role.
• Context
  • The AI needs to know the context for what it is asked to do; it’s a machine, so it can’t derive it naturally.

3.2 Given an initial prompt and its output, identify additional inputs you can use to elicit more targeted output.

• Examples (few-shot prompting).
• Glossary for translation.
• Templates.
• Documents to use for research.
• Earlier conversation in the same thread.

3.3 Recognize common prompting techniques.

• Zero-shot, few-shot, chain-of-thought, self-consistency, generate knowledge, prompt chaining.

3.4 Use reverse prompting techniques to achieve an outcome.

3.5 Given an AI output, explain how you can verify the accuracy of the output.

• Historical facts.
• Current facts.

Ethics, Law, and Societal Impact - 21 questions

4.1 Identify the potential for bias in Generative AI output.

• AI can reflect the biases present in its training data.
• Different models might have different biases.
• The creator of the model introduces bias by adding guardrails.
  • Some tools allow their additional guardrails to be turned on and off (Azure Open AI for example)
• Bias can be introduced through the prompt.
• Common biases include gender, race, disability, age, religion, cultural, language, nationality, and economic status.
• Generative AI can be used to propagate bias.

4.2 Identify the potential legal implications of using Generative AI.

• Honoring intellectual property rights
  • The laws are still in flux however, the best approach is to use non-AI practices and not use another person’s work without permission.
  • Copyrighted data was used to train AI for some models.
• Identifying legal implications of inappropriate use of generated content.
• Transparency – documenting your process when using AI in a professional environment.

4.3 Explain the importance of data privacy.

• Personal information or a company’s private data could be used for training.
• Identity theft might occur if personally identifiable information (PII) is used by Generative AI.
• Identity theft could result in civil and criminal actions.
• Companies are establishing internal policies to prevent employees from leaking data to public/unapproved AI models.
• Human-generated content might be used to train the model unless you opt out.

4.4 Determine the risks associated with using Generative AI.

• Necessity of human oversight to avoid spreading incorrect or harmful information that leaves you or the company vulnerable to financial and/or legal repercussions.
• Understanding that you are responsible for what you create.
  • Refraining from creating content that is harmful or could potentially lead to civil or criminal actions (bullying, hate crimes, fraud, stalking, cheating)
  • Generative AI can be used for dangerous purposes, including deep fakes; easier to generate harmful or illegal information that looks real; identity theft.

4.5 Identify the impacts of Generative AI on society.

• Negative
  • Recognizing the implications of the reduction of human interaction
  • Recognizing that AI does not replace human contact
  • Recognizing the possible impact on human motivation due to overreliance on AI
  • Recognizing the human motivation to use AI to sway public opinion
  • Fear that AI will take over our jobs and our humanity
  • Socioeconomic factors – AI is not available to everyone equally
• Positive
  • Generative AI can help us do our job more efficiently.
  • Help us communicate better, particularly across languages.
  • Help us learn more effectively.
  • Generate ideas to spark creativity; brainstorming.
  • Help us do our life tasks more efficiently – menus, recipes, grocery list, summarize long messages from friends and family
  • Analyze patterns and presenting them as opportunities
  • Create jobs – they will just be different jobs