Generative AI Flashcards
(106 cards)
1
Q
What is one real-world application of Variational Autoencoders (VAE) in anomaly detection?
A
Detecting defects in industrial quality control by identifying images of products that deviate from a dataset of normal products.
2
Q
What are Midjourney, DALL-E, and Stable Diffusion, and which industries are their early adopters?
A
They are primary text-to-image generation services and models. Art, filmmaking, fashion, and marketing are the first industries to widely adopt their use.
3
Q
What is GPT and why has it become notable in the field of natural language processing?
A
GPT is a language model developed by OpenAI that can take in a prompt and generate text based on it, making it useful for a multitude of tasks.
4
Q
How does a GAN network improve its ability to generate better content?
A
The generator and discriminator parts of the network work together in a competition to improve the generator’s ability to create realistic data.
5
Q
What is the purpose of the discriminator in a GAN model?
A
to evaluate the data created by the generator and give feedback on how to improve the next iteration
6
Q
What is a key benefit of using generative AI for repetitive or computational tasks?
A
It allows humans to focus on more creative and strategic activities.
7
Q
What is the difference between generative AI and other types of AI that generate content?
A
Generative AI’s primary function is to create content.
New content that AI generates can be in the form of text, images, or even product suggestions.
8
Q
What does the term open source mean in the context of generative AI models?
A
It means that the models are publicly available for anyone to use and modify.
9
Q
What is an AI model?
A
A model is a set of algorithms that have been trained on a specific dataset.
10
Q
What will be the main benefit of generative AI in the next years?
A
automate repetitive tasks and liberate humanity from dull, dirty, difficult, or dangerous jobs
11
Q
What are the top moral and executive skill sets required when working with generative AI?
A
Transparency, fairness, empathy and responsibility. Approach production and operations with caution, always asking, “Who is benefiting?” from our generative AI solution.
12
Q
When considering the integration of generative AI tools in business operations, what is the primary emphasis regarding the role of executive leadership and organizational strategy?
A
prioritizing human-centered approaches, ethical considerations, and maintaining human control over AI-generated content
13
Q
What is the goal of the continuous crawling process of a search engine?
A
to keep the search engine’s index up-to-date
14
Q
When a user enters a query, what does the reasoning engine strive to provide?
A
a relevant, informative text response using human-like speech
15
Q
When might a search engine be a superior好過其他的 tool to a reasoning engine?
A
when you’d like to read further about a subject across a collection of different sources—but not necessarily when you want to ask deeper questions
16
Q
Which is not a main function of a search engine?
A
transforming
The main functions of a search engine are crawling, indexing, and ranking.
17
Q
What is the most important benefit that the synergy協同作用 between modern search engines and reasoning engines provides, as far as confidence in the results?
A
verifying and validating search results
18
Q
How can a user best combine a search engine and a reasoning engine to find information about an unknown topic?
A
Use the search engine to find basic information, and then use the reasoning engine for a deeper dive.
19
Q
How does a reasoning engine’s ability to understand and interpret language provide the greatest advantage over a search engine?
A
It can have an actual conversation with the user.
20
Q
How are reasoning engines an improvement over search engines when it comes to entering what you are looking for?
A
They can understand your intent and not just the words you used.
21
Q
How do human supervisors assist in training a reasoning engine?
A
In early training phases, human supervisors oversee the process, guiding the model towards accurate responses and contributing to its knowledge development.
22
Q
True or False: Reasoning engines are an all-knowing source of truth and should be trusted implicitly.
A
FALSE
23
Q
When a user enters a query, what does the reasoning engine strive to provide?
A
a relevant, informative text response using human-like speech
24
Q
When might a search engine be a superior tool to a reasoning engine?
A
when you’d like to read further about a subject across a collection of different sources—but not necessarily when you want to ask deeper questions
25
If your reasoning engine response is problematic (i.e., inaccurate, discriminatory, limited in view, etc.) what should you do?
Continue iterating. Keep regenerating and refining the prompt to get a more accurate, better result.
26
In prompt engineering, what is one-shot or few-shot learning?
It refers to how much instruction you provide in order to guide the answer. This may involve including examples of what a "correct" answer may look like.
27
In most instances, how should you craft your prompts?
Use clear language with proper grammar.
28
Why is the iteration process necessary when you use a reasoning engine?
You want to keep honing your prompt to get more and better results.
29
Riva considers herself a prompt engineer. What does this mean?
She can create a prompt with samples of her question and the answers that she would like to see.
30
What is the following creative type of prompt known as: “Imagine you’re the manager of a small botique精品店 video editing company. What are 10 innovative marketing ideas that could attract new business?"
Role play
31
True or False: Reasoning engines are an all-knowing source of truth and should be trusted implicitly.含蓄地
FALSE
32
Excel Copilot can only be used if
the file you are working on is stored on OneDrive, SharePoint, or Teams
33
the file you are working on is stored on OneDrive, SharePoint, or Teams
the cells you want to analyze are formatted as a table
34
If you join a Teams meeting that's already in progress, you can ask Copilot to summarize the topics that were discussed that you may have missed. This will only work if
somebody in the meeting clicks the Copilot option and turns on the transcript feature
35
PowerPoint Copilot can create a new presentation based on your written prompts or
information in an existing document
36
What is a common reason the summary option may not be available in Outlook?
The message you have selected is very short and does not have enough information to summarize.
37
If you want to ask Word Copilot to use information from a separate document, that referenced document must be
shared in Teams or stored on OneDrive, SharePoint
38
Damien wants to know who the target population is for the AI tool and what their main goals are. Which pillar in the AI framework is he addressing?
boundaries on safe and appropriate use
39
Cuong is auditing the company's new AI tool. He determines that customers are using it beyond what it was intended for. What should be his first step when handling this issue?
Take it offline.
40
What tools can be used to create a persona to deliver fraudulent欺詐的 information?
deep fakes
41
Which pillar of the ethical AI framework is the starting point for all ethical AI tools?
responsible data practice
42
Which responsibility falls into the realm 領域of the board?
哪些責任屬於董事會的範疇
Ensure the company has resources to manage AI ethical risks.
43
As a chief AI officer, Natasha is reviewing with one of the teams the purpose of the product and possible risks to users. Where does this task fit into the LISA framework for listening to customers?
Auditing
44
In which of the three goals for an effective and ethical data organization would a training curriculum on data responsibility be promoted?
prioritizing privacy
Everyone must understand how to handle sensitive data and keep it secure or risk losing trust.
45
Lately, a technical team has been facing several challenges in the use of technology at the company. Which action should the company and the team take first?
Encourage everyone to ask questions and raise concerns.
46
In using the acronym ETHICS to adhere to responsibilities for yourself and those around you, which letter represents the group responsible for feedback and insights?
C
- I: These are the industry experts who have a responsibility to share their knowledge and experience on the ethical implications of AI.
- T is for technologists, engineers and developers, who must ensure that AI systems are secure, safe, and compatible with ethical frameworks.
- H represents human rights advocates who have a responsibility to ensure that technologists respect human rights when building systems.
47
A large company has been trying to implement AI. The C-suite established an AI policy and governance framework. What should the company do to sustain its AI efforts?
Establish accountability for AI ethics implementation in a C-Suite role, which might be a chief AI ethics officer.
48
Why did some of the earliest artificial intelligence systems focus on board games such as checkers and chess?
It's easiest to make a computer system seem intelligent when it's working with set rules and patterns.
49
You're a product manager who's in charge of building a weak AI expert system that will give tax advice. You're working with dozens of accountants who go through thousands of different taxpayer scenarios. When a customer asks a question, then the expert system will ask a follow-up question. It will do this until it makes a recommendation. What's one of the biggest challenges with this system?
There will be too many tax combinations for the experts to cover with one system.
50
Luella seeks medical attention for chest pains. A nurse uses an artificial intelligence program to diagnose the cause. Why is this system likely not really intelligent?
The program only matches her symptoms to steps in a system an expert created.
51
How does an artificial neural network learn?
t looks at the data and makes guesses, then it compares those guesses to the correct answer.
52
The healthcare and medical insurance industries caution against using machine learning to search for patterns in data, and they do not want machines making decisions about a person's health. Why?
They may be decisions that humans cannot understand.
53
What type of impact does artificial intelligence have on robotics?
AI systems can create robots that can more easily learn new tasks.
54
What impact will the Internet of Things (IoT) have on artificial intelligence?
These devices will be a great new source of “real world” data.
55
A new online camping goods store wants to find connections between products customers buy and other products they might buy. Why would the company use unsupervised learning?
It does not yet have enough customers to make supervised learning meaningful.
56
You're a preschool worker and you want to teach your class the letters in the alphabet. So you draw the letter “B” on the board. Then you ask the two-year-old students to find a block with that same letter. Some of the students correctly find the blocks with the letter “B”, but some of the students confuse the letter “B” with the letter “D.” So the incorrect students compare their block to the letter “B” on the board, recognize the error and then decide to get another block. What type of learning is this?
supervised learning
Supervised learning relies on labeled data. In this case the two-year-old students would use the letter written on the board as labeled data. Then they would try to classify the unknown data (the blocks) by matching the label. If the student sees that they made a mistake, they adjust and take another guess.
57
Why might you want to use reinforcement learning instead of unsupervised learning?
Reinforcement learning allows the machine to make predictions and create strategies instead of just clustering the data.
58
What is one of the greatest challenges with supervised learning binary classification?
You need a lot of pre-classified or labeled data for the training set.
59
You work for a company that’s selling electric cars to consumers. The company wants to get the maximum amount of value from its advertising dollars. So it wants to ramp up advertising when it thinks that customers would be most interested in purchasing an electric car. Your data science team wants to create a regression analysis based on fuel prices. How might this look on an XY diagram?
Create a trendline with fuel prices along the X axis and electric car sales on the Y axis.
To do a regression analysis, data science teams will typically plot their data on an XY diagram. Then they will see if there are any trends in the data by creating a line in the center of the largest data point groupings. This is typically called a trendline. If the data has a clear trendline then it will be easier to predict relationships between the two variables. In this case, there might be a clear trend where when fuel prices rise, people are more likely to buy electric cars.
60
How is K Nearest Neighbor like the old saying, “birds of a feather flock together?”物以類聚，人以群分
Classify unknown data against the closest data that you do know.
61
What is ensemble modeling?
This is when you use a mix of different machine learning algorithms or data to improve the outcome.
62
You work for a credit card company that's trying to do a better job identifying fraudulent transactions. So your team uses unsupervised learning to create clusters of transactions that are likely to be fraudulent. The machine identified that when customers are buying electronics it's much more likely to be a fraudulent transaction. So you use this model for your new fraud detection system. Then customers started to complain that they couldn't use their credit cards to purchase any electronics. What is the challenge with your model?
You underfit the model to the data, the simple rule made too many inaccurate predictions.
Sometimes you can identify patterns that work with a small set of data but that doesn't fit when you start to look at larger datasets. This is called underfitting the data. Other times you can add more variables. This can create a lot of complexity and you might miss data outliers (data that is close but doesn't quite fit the model). This is called overfitting the data.
63
How does the bias-variance trade-off 協調affect machine learning?
If the machine makes a change to one, it must consider how the other is affected.
64
Kira is building a neural network to identify customer returns using binary classifications of defective or unsatisfied. In which layer of this neural network will Kira have a probability score?
the output layer
65
You work for a security firm that wants to use an artificial neural network to create a video facial recognition system. So you create a training set with hundreds of images of people that are found in your video footage. You initialize the artificial neural network with random weights assigned to all its connections. When you feed through the first few images the system does a terrible job identifying whether those people are included in the video. What would the artificial neural network now do to try and improve?
It will adjust the weights of the connections to see if it does a better job making a prediction.
66
With an artificial neural network what is the point of having a cost function?
It helps the network determine the cost of the error so they can make larger or smaller adjustments to its guesses.
Artificial neural networks need a measurement of “wrongness.” That way it knows how much to adjust its weights and biases. This is typically done through a calculation of the gradient傾斜度 descent which will increase or decrease the cost function. If it's very wrong, it will make big changes to the weights and biases. If it's slightly wrong, it will make much smaller changes.
67
How can you best describe the cost function as it applies to neural networks?
a number the system uses to measure its answer against the correct answer
68
You are an executive for a large company that has a customer service department. Recently some of the top managers have been talking about replacing customer service representatives with an AI chatbot. Some of the managers feel like the chatbot should impersonate a human customer service representative. They argue that if customers know it’s an AI chatbot then they would immediately disconnect. Other managers feel like it would be unethical to impersonate a human. What would be the best place to communicate your decision?
Create a Responsible AI Policy and Governance framework.
69
You’re a director for an organization that detects credit card data. You’re trying to convince your manager to adopt a generative adversarial network (GAN) to test your system to see if it can identify credit card fraud. What’s one of the best arguments you have for using this type of neural network?
Fraudulent transactions are by their very nature adversaria對立的l, so it’s good to have a network that reflects this.
70
What is a good description of how a machine learning system operates?
A system “learns” by observing patterns in massive datasets.
71
Oxford dictionary defines plagiarism抄襲 as “the practice of taking someone else’s work or ideas and passing them off as one’s own.” If you ask ChatGPT to describe a sunset, it will give you a response, but these systems have never experienced a sunset. The only way it could respond is by “passing off ideas as its own.” Does that mean that these generative AI systems are plagiarism machines?
No, these systems may be thought of as experiencing events that it hasn’t experienced.
72
You are an executive of a company that is implementing generative AI systems. What are the most essential ethical considerations to balance?
your organization’s obligation to appease shareholders against your obligations to humanity
73
A national newspaper reporter is writing a story on generative AI. As part of the story, they chat for hours with a new online generative chatbot. A few hours into the conversation, the chatbot tries to convince the reporter to leave his partner. The chatbot company said they don’t know why it gave these responses and will limit conversations to 30 minutes. What might be one of the biggest ethical challenges with this system?
There isn’t enough transparency into how the chatbot is responding.
74
Why did early artificial intelligence systems do so well with board games?
Because even with their limiting processing power, early systems thrived in a world of simple rules and pattern matching.
75
What does the term model mean in generative AI?
A model is a set of algorithms that have been trained on a data set.
76
You’re trying to get better at prompt engineering, so you decided to try a new technique. You say, “Write a 500-word essay on large language models and hallucinations 幻覺from the perspective of a computer science graduate student at a university.” What technique are you using here?
You are using role-playing to get more accurate responses.
77
In machine learning, when a data model performs exceptionally well during the training set phase, but lacks the complexity to generate accurate predictions during the test set phase, the model is _____ the data.
overfitting
78
You’re trying to improve your skills with prompt engineering, so you asked ChatGPT to generate a paragraph of text. The first prompt you create is, “Tell me about lactose intolerance.” You weren’t satisfied with the results, so for the second prompt you wrote, “Write a blog article on lactose intolerance for my healthcare website.” What did you do with the second prompt that you didn’t do with the first?
You provided context.
79
How is an artificial neural network neural network related to machine learning?
An artificial neural network is a machine learning technique.
80
What is a generative adversarial network (GAN)?
when two neural networks work in opposition, with a generator and a discriminator to improve the generative output
81
You work for a large financial institution that wants to identify undervalued stocks存貨. To do so, you feed decades of financial information into an artificial neural network neural network to create clusters of stocks. Then your data science team tries to find stocks in those clusters that substantially increased in value. Your data science team hopes to find stocks in the same cluster that may also gain value. What type of machine learning are you using?
unsupervised learning
82
Your large social media company has decided to open source the data and source code for your chatbot. You recently found out that a foreign government has downloaded your code and set up a chatbot to spread propaganda. The chatbot encourages violence 激烈的言辭against an ethnic minority group. What AI ethics violation might your chatbot release have caused?
Your technology assisted a human rights violation
83
You work in the marketing department for a large company, and you’d like to create a weekly opinion letter using ChatGPT to give your take on the top news in your industry. You create a few test posts, and you notice that ChatGPT is getting the dates wrong and is mixing up the CEOs of different companies. Why are you running into this challenge?
ChatGPT shouldn’t be used for creative writing because it’s still prone to factual errors.
84
You are having some difficulty dealing with a colleague at work. You asked ChatGPT for advice on how to improve your relationship with the coworker. ChatGPT gives you extremely helpful advice. You find yourself intuitively thanking it for its help. Given your interaction, do you think that ChatGPT is strong or weak AI?
It is weak AI because ChatGPT doesn’t understand what it’s saying—it’s just gathering information that it found online.
85
Your company produces science fiction and fantasy graphic novels. One of your top illustrators has developed a style that is very strongly associated with your brand. Your company decides to create a generative AI model to mimic their illustrations模仿他們的插圖. This new model can create new graphics in their style in seconds. Now the company will have better control over their brand and increase productivity. What is one of the main challenges with this approach?
It will “normalize mediocrity”—the graphics will look the same and lack a creative spark.
86
Your online movie-streaming business wants to create an artificial neural network neural network that can recommend new movies based on what customers have already seen. The team creates a series of XY diagrams of different film genres. Then it puts the film rating along the X-axis and the duration that people watch on the Y-axis. It then makes a recommendation based on how close movies are to each other on the chart. What type of machine learning algorithm is the team using?
K-nearest neighbor
87
You are a technical manager for a large city courthouse. The judges have asked you to implement a new system that will make criminal sentencing recommendations. As part of your testing, your team has the system make sentencing recommendations for past court convictions. Your team finds that the new system is much more likely to recommend longer sentences for some groups of people. What is the main ethical challenge with implementing this system?
It magnifies existing biases rather than mitigating them.
88
What is the difference between generative AI and discriminative AI?
Generative AI creates content while discriminative AI classifies data.
89
You work for a political organization that does sentiment analysis of social media networks. Politicians look to your service to see how people feel about certain difficult topics. Your organization has developed an artificial neural network neural network that can search social media for topics and classify the comments as strongly agree, neutral, and strongly disagree. What type of machine learning are you using?
supervised learning multiclass classification
90
You are a software developer on a team that’s developing a generative AI nurse for a healthcare company. You’ve trained the system on all your internal data, but to make it more “worldly” you’ve also trained it with social media data. During your testing, you found that sometimes the nurse will make recommendations that aren’t based on science. As a software developer, your AI ethical responsibility is to make sure that the AI nurse _____.
is developed in a way that’s transparent, explainable, and accountable
91
You work for a large credit card company that wants to create an artificial neural network neural network that will help predict when people are going to have trouble paying their bills. So your team gathers all the billing statements for people who had trouble paying their bills. Then you feed this data into an artificial neural network neural network. What is this process called?
This is training your artificial neural network with labeled data.
92
Your company wants to use generative AI to come up with new pharmaceuticals. This system will analyze all existing chemical compounds and try to develop new compounds based on the success of some of your current pharmaceuticals. This system will require a lot of custom programming and access to your proprietary data sets. What type of generative AI system might work best?
Develop your own generative AI model based on your existing data.
93
You are going to use machine learning to try and do a better job predicting the weather. To start out, you just want to classify two weather events: “rain” or “not rain.” What steps would you take to build this system?
Find labeled weather data, create a small training set of that data, and that set aside more data for the test set.
94
You’re an executive for a software development company. Your company develops only one product. You want to include ethical decision-making into your software development, so you ask a senior developer to also serve as the company’s chief AI ethics officer. What would be one of the challenges with this approach?
A chief AI ethics officer sets the ethical direction for the entire company and shouldn’t just focus on the product.
95
You recently purchased a new smartwatch. To set up the watch, you had to go to the manufacturer’s website and create a new account. When you create an account, it presents a long license agreement you have to accept to create an account. You were anxious to use your new watch, so you didn’t scroll through the 50-plus pages of the license agreement. What is one of the ethical issues with how the smartwatch manufacturer is operating?
They are not sharing a clear and transparent privacy policy.
96
What’s one of the key dangers for organizations that over rely on generative AI systems?
They will regenerate the same material without any spark of creativity.
97
You work for a large financial institution that would like to offer immediate approval for loan applications. Your team has identified four predictors about whether someone will be a good loan candidate: income, credit score, employment, and debt. You develop a system that will look at each predictor independently and then come up with an overall score. What machine learning algorithm are you using?
Linear regression
98
You work for a company that wants to improve spam filtering for mobile email applications. Your data science team gathers one million messages that have been correctly labeled as spam. You then train an artificial neural network neural network to correctly identify these spam messages. After you train the system, one of the product managers asks why you don’t use those same million messages to test the network for accuracy. How should you respond?
If you use the training data, then you’re not testing how well the system will do in the future to identify spam.
99
How does a reasoning engine work?
It draws conclusions, makes decisions, summarizes information, and solves problems based on available data.
100
You use a new text-to-image generating service to create a beautiful artistic landscape. You submit your artwork for an AI-generated artistic award and win third place. The news media picks up the story and uses your image in online articles without any compensation or attribution. Did they violate your copyright protection?
Yes, your image and the prompt engineering phrase can be protected by copyright.
101
When you’re approached with a generative AI ethics challenge, what is one of the first questions you should ask?
What is the highest standard of responsible human behavior?
102
You work on a team that’s developing a generative AI text-to-image service. Your service will specialize in creating realistic looking paintings. To train the system you have to process millions of digitized paintings. The system has learned that paintings almost always have a signature. When you test the system, it creates a fake signature on the painting. The product manager asks you to create an algorithm to remove the signatures. What might be an ethical challenge with this approach?
It should allow the system to create a fake form of attribution.
103
Typically, what are the three layers of an artificial neural network neural network?
the input layer, many hidden layers, and the output layer
104
You work for a company that produces video games. One of the challenges is creating non-player characters (NPCs) that are controlled by the game, but still make strategic decisions. Your team decides to use machine learning, and each time an NPC does better than a player it gets a small reward. Now the machine learning algorithms are coming up with interesting new ways to play the game. What type of learning is this?
reinforcement鞏固 learning
105
Your company wants to create a smartphone application that identifies plants using the phone’s camera. The company purchases millions of digital images of plants labeled with the species names. You use this initial batch of images to train your artificial neural network neural network. What type of machine learning are you using for your network?
supervised learning
106
You manage a radiology department in a large hospital. Your hospital has millions of computed tomography (CT) images. You want to create a system where once someone gets a CT scan, the system will immediately check for anomalies. That way it can be sent for review by a senior radiologist. Which generative AI system might work best for this approach?
a generative autoencoding network (GAN)
