Create a new flashcard set

M1 | What is generative AI?

AI that learns patterns from data and produces new content such as text, images, audio, video, or code.

M1 | How does generative AI differ from a traditional classifier?

A classifier assigns a label or predicts a value; generative AI synthesizes new content.

M1 | What are the four nested concepts in the module?

AI is the broad field; ML learns from data; DL uses multilayer neural networks; generative AI is presented as a DL application that creates content.

M1 | A spam filter learns from examples to label emails. Which category best describes it?

Machine learning for classification.

M1 | Facial recognition using multilayer neural networks is an example of what?

Deep learning.

M1 | A system writes a new essay from a prompt. Which category best describes its task?

Generative AI.

M1 | What does data-driven learning mean for generative AI?

Learning patterns from large datasets to generate plausible new outputs.

M1 | What is multimodality?

Working with or generating more than one type of content, such as text, images, audio, and video.

M1 | What does human-like interaction mean in generative AI?

Producing conversational responses that resemble natural dialogue; it does not imply human understanding.

M1 | What is creativity or novelty in generative AI?

Producing a new combination or output, such as artwork, product concepts, or marketing copy, from learned patterns.

M1 | What model family is associated with realistic synthetic images and deepfakes?

Generative adversarial networks (GANs).

M1 | Which model family is associated with text-to-image generation such as Stable Diffusion?

Diffusion models.

M1 | Which model family is chiefly used for generating essays and code in the module?

Large language models (LLMs).

M1 | What distinguished ELIZA (1966) from modern data-trained chatbots?

ELIZA used rule-based pattern matching to imitate conversation.

M1 | What did expert systems of the 1980s rely on?

Rules and specialized domain knowledge, for example in medicine and engineering.

M1 | What shifted in the 1990s with machine learning?

Systems increasingly learned patterns from data instead of relying solely on fixed rules.

M1 | What enabled major deep learning progress in the 2010s?

Neural networks trained with substantial data and GPU computation.

M1 | What year does the module associate with GANs, and what did they advance?

2014; realistic image synthesis.

M1 | Which era does the module associate with diffusion models becoming prominent?

The 2020s, especially for text-to-image generation.

M1 | Who posed the question "Can machines think?" and who coined "artificial intelligence"?

Alan Turing posed the question in 1950; John McCarthy coined the term in 1956.

M1 | Name one application of generative AI in healthcare.

Designing candidate molecules for drug discovery; proposed compounds still need scientific validation.

M1 | Name one application of generative AI in education.

An AI tutor that helps personalize explanations and guide problem solving.

M1 | Name one application of generative AI in creative work.

Generating concept art, helping edit video, or drafting creative material.

M1 | What three ethical issues does Module 1 associate with generated content?

Bias, authenticity or misinformation such as deepfakes, and intellectual property.

M1 | A realistic AI video of a politician circulates. Which capability and concern are illustrated?

Synthetic video generation and the risk of misinformation or deceptive deepfakes.

M2 | How does an LLM generate a response at a simplified level?

It uses learned patterns and prompt context to predict and produce text token by token.

M2 | Does an LLM normally retrieve an exact stored answer whenever it replies?

No. It generates a response from learned patterns; a separately connected search or retrieval tool can supply sources.

M2 | What do "Large," "Language," and "Model" mean in LLM?

Large: trained on huge data; Language: handles human language; Model: a mathematical system of learned patterns.

M2 | Why can an LLM sound intelligent without thinking like a person?

It is effective at predicting contextually plausible language, but this is not human understanding or emotion.

M2 | What three simplified stages describe generative AI in Module 2?

Training data → learning patterns → generating a response from a prompt.

M2 | What is a token?

A small unit of text a language model processes; a word may be one or several tokens.

M2 | Why do tokens matter?

They affect input and output length, context capacity, and processing cost.

M2 | What is conversational context?

Earlier information available in the conversation that helps interpret later messages.

M2 | What is a context window?

The limited amount of input and prior conversation, measured in tokens, the model can consider at once.

M2 | Maria gives her name earlier in a chat, then asks for it. Why can the model answer?

Her name remains in the available conversation context.

M2 | Maria starts a new chat and asks for her name without supplying it. What should the model say?

It cannot know from that new chat alone; it should ask or say it lacks the information.

M2 | What is prompt engineering?

Designing instructions and context to obtain a more useful response.

M2 | What four components does the module recommend for a clear prompt?

Role, task, audience, and output format.

M2 | A prompt says "Explain databases." What change makes it more useful for a first-year student?

Specify the audience, a familiar analogy such as a library, and the desired examples or format.

M2 | What is zero-shot prompting?

Asking the model to perform a task without supplying examples.

M2 | What is few-shot prompting?

Providing a small number of examples to illustrate the desired task or output pattern.

M2 | When is few-shot prompting especially useful?

When the desired classification or format is easier to demonstrate with labeled examples.

M2 | What is role prompting?

Assigning a role or perspective, such as "Act as a professor," to guide the response style and focus.

M2 | A student wants a beginner explanation and short quiz. What should the prompt specify?

Topic, beginner audience, simple language or analogy, and a quiz with the desired number or format.

M2 | How can ChatGPT help with studying?

Summarize, simplify, create practice questions, and draft study guides; verify learning against the source.

M2 | How should a student treat AI-generated programming code?

Review it, test it, and check requirements; ask for input validation and error handling where appropriate.

M2 | Name two professional content tasks suited to drafting with ChatGPT.

For example, drafting an email and a product description or social media caption.

M2 | What is an AI hallucination?

A confident-sounding answer that includes false or fabricated information, such as a nonexistent citation.

M2 | Why can an answer sound certain but be wrong?

Language plausibility and confidence of wording do not guarantee factual verification.

M2 | Which five checks does Module 2 recommend for AI responses?

Accuracy, sources, logic, relevance, and bias.

M2 | Why should recent facts be checked outside an LLM response?

Knowledge may be outdated, and the model may generate plausible but inaccurate claims.

M2 | Give two appropriate academic uses and one inappropriate use of ChatGPT.

Appropriate: studying and brainstorming or drafting; inappropriate: submitting unverified AI work as one's own in violation of course rules.

M2 | What does changing a prompt from "Explain AI to a Grade 6 student" to "Explain AI to a CS student" demonstrate?

Audience and context change the depth, vocabulary, and examples of the output.

M3 | Define Responsible AI.

Ethical design, development, and use of AI so it is fair, reliable, safe, private, inclusive, transparent, and accountable.

M3 | Name the six Responsible AI principles in the module.

Fairness; reliability and safety; privacy and security; inclusiveness; transparency; accountability.

M3 | How do transparency and accountability differ?

Transparency helps people know about and understand AI use; accountability identifies who must answer for outcomes.

M3 | What is AI bias?

Unfair or discriminatory outcomes associated with training data, design choices, or the setting where AI is used.

M3 | A face model trained mostly on light-skinned people performs worse on darker-skinned people. Which bias source is illustrated?

Data bias: the training data does not adequately represent the people served.

M3 | A hiring model uses home address as a proxy for income. Which bias source is illustrated?

Design bias: chosen features can indirectly disadvantage a group.

M3 | An English-only chatbot serves Filipino-speaking users. Which bias source is illustrated?

Deployment bias: it is used with people or circumstances it was not designed or tested for.

M3 | What did the Amazon recruiting tool illustrate?

Historical résumé data can encode gender discrimination and reproduce it in hiring recommendations.

M3 | What did the Gender Shades study illustrate?

Facial-analysis error rates can differ greatly across intersectional groups; test performance by group.

M3 | What did the COMPAS controversy illustrate?

Unequal errors in criminal-risk assessment can affect people's liberty.

M3 | How can you test a generated story for stereotypes?

Look for unrequested assumptions about gender, age, class, or nationality; revise the prompt and compare outputs.

M3 | How can a prompt for sample applicants reduce stereotyping?

Vary demographic backgrounds deliberately and state that skill must not be inferred from them.

M3 | What is fairness in AI?

Seeking equitable outcomes and avoiding disadvantage based on protected or relevant group characteristics.

M3 | What three fairness checks does the module distinguish?

Compare outcomes or approval rates, error rates, and opportunities for qualified people across groups.

M3 | Why can removing a sensitive attribute fail to remove discrimination?

Other variables such as address or school can act as proxies.

M3 | What is a hallucinated citation?

A reference whose author, title, journal, DOI, or existence is fabricated or misrepresented.

M3 | What causes hallucinations according to the module?

Next-token prediction without automatic fact checking, information gaps or stale knowledge, and pressure to provide an answer.

M3 | How should you verify an academic source suggested by AI?

Search the title in a scholarly database, check its DOI or publisher record, and compare the actual paper with the claim.

M3 | What do "real," "partially correct," and "fabricated" mean in a citation audit?

Real: details match a source; partially correct: a source exists but some details are wrong; fabricated: no matching source exists.

M3 | What is Retrieval-Augmented Generation (RAG)?

A method that retrieves relevant trusted documents and supplies them as context before generating an answer.

M3 | What are the four simplified RAG stages?

Question → retrieve documents → add documents to context → generate a grounded answer.

M3 | Does RAG remove the need to verify citations?

No. Open the cited sources and check whether they support each claim.

M3 | What does Mata v. Avianca illustrate?

Lawyers cited nonexistent AI-generated court cases and were sanctioned; professionals must verify sources.

M3 | What does Moffatt v. Air Canada illustrate?

An organization may be held responsible for incorrect advice provided by its chatbot.

M3 | Define explainability in AI.

Being able to understand and interpret the basis for an AI decision, especially when the decision affects a person.

M3 | Compare black-box and glass-box models.

Black-box models are hard to interpret internally; glass-box models such as simple rules or scorecards expose their decision steps.

M3 | What do SHAP and LIME aim to explain in the module?

Which input features contributed to a particular model prediction; explanations still require careful interpretation.

M3 | What is a counterfactual explanation?

A statement describing a plausible change to an input that would change the model's decision.

M3 | What is a model card?

Documentation of a model's intended use, limits, and evaluation results.

M3 | Why is explainability particularly important in healthcare, lending, and scholarship selection?

People need to inspect, challenge, and correct consequential decisions and identify errors or bias.

M3 | Who should make the final decision when AI ranks scholarship applicants?

Accountable people under the institution's review process, informed by checked evidence and fairness/privacy safeguards.

M3 | What is AI governance?

Policies, laws, standards, and oversight guiding responsible AI development and use.

M3 | Name the three governance levels and one example of each.

Organizational: school AI policy; national/regional: Philippine Data Privacy Act or EU AI Act; global: UNESCO or OECD principles.

M3 | What is the Philippine Data Privacy Act called, and who enforces it?

Republic Act 10173, the Data Privacy Act of 2012; the National Privacy Commission (NPC).

M3 | What is NPC Advisory 2024-04 about in the module?

Guidance on applying data privacy rules to AI systems that process personal data.

M3 | Why was the Samsung ChatGPT incident used as a privacy example?

Employees pasted confidential code and meeting content into a chatbot, taking sensitive data outside company control.

M3 | What should you do before putting a student's grades into an AI chatbot?

Avoid exposing them unless authorized; establish identity and access controls, a lawful basis and clear privacy safeguards under applicable policy and law.

M3 | Which information should generally not be pasted into a public AI chatbot?

Student records, IDs, passwords, API keys, confidential work documents, and other people's private messages or photos.

M3 | What are safer alternatives to pasting private records?

Remove identifying details, summarize the situation, use fictitious examples, or ask about the general concept.

M3 | What two copyright questions does generative AI raise?

Whether copyrighted works may be used for training and who can claim rights in AI-generated output.

M3 | What was alleged in Getty Images v. Stability AI and The New York Times v. OpenAI/Microsoft?

Unauthorized use of copyrighted images and news articles, respectively, for AI training; these are allegations, not automatic findings.

M3 | What distinction does the module draw using "Zarya of the Dawn"?

The human-written text and arrangement were protected; images generated by Midjourney alone were not protected in the cited U.S. decision.

M3 | What does Thaler v. Perlmutter illustrate in the module?

Under the cited U.S. ruling, a work created entirely by AI without human authorship cannot receive copyright protection.

M3 | How should a student disclose AI help in an assignment?

Follow the syllabus; say what AI helped with and what research, verification, and final writing the student performed.

M3 | When should a student cite ChatGPT, according to the module?

When quoting or relying on its output, subject to the course's citation rules.

M3 | What is the EU AI Act's main classification approach?

It groups AI uses by risk; obligations increase as risk rises.

M3 | List the four EU AI Act risk categories.

Unacceptable, high, limited, and minimal risk.