CAI 3801 Week 2 Study Notes

CAI 3801 • Week 2 GenAI Literacy for Business

  • Course Overview:

    • Transition from understanding GenAI as a 'cool tool' to utilizing it for professional business deliverables.

    • Goals for this Session:

    • Understanding what Generative AI (GenAI) is particularly good at.

    • Conceptual understanding of how Large Language Models (LLMs) function.

    • When it is appropriate to trust GenAI outputs.

    • Methods to make GenAI outputs usable for business.

    • Accountability Focus: Context, evidence, and decisions.

Transition from Week 1 to Week 2

  • Recap of Week 1:

    • Emphasized how analytics supports human decision-making.

    • Defined AI's function as learning from data and acting (A → B).

    • Introduced Machine Learning (ML) and Deep Learning (DL).

    • Explained the concept of LLMs generating text through prediction.

  • Focus for Week 2:

    • Developing GenAI literacy: understanding its abilities and limitations.

    • Overview of three task families GenAI addresses.

    • Strategies for transforming GenAI outputs into effective business deliverables.

  • Preview of Week 3:

    • Exploring prompting as a method of managerial reasoning.

    • Topics will include writing strong briefs, controlling output formats, and formulating follow-up questions.

    • Course promise emphasizes practical judgment over mere AI tricks.

Understanding AI's Presence in Daily Life

  • Common AI Applications:

    • AI is already integrated into daily life, though often not referred to explicitly as "AI.">

    • Examples:

    • Netflix recommendation systems that suggest content based on user behavior.

    • Amazon predicts purchase behavior and recommends products before users explicitly search.

    • AR filters on social media platforms like Snapchat that adjust based on facial recognition.

    • TikTok's For You Page employs algorithms to retain user engagement.

Demystifying AI

  • Types of AI:

    • Generative AI:

    • Focuses on creating new content such as text, images, music, and more.

    • Sub-field that has gained traction in commercial applications.

    • ANI (Artificial Narrow Intelligence): AI specialized in one task to outperform humans in that area.

    • AGI (Artificial General Intelligence): Hypothetical AI possessing reasoning, learning, and comprehension abilities akin to a human.

Questions Regarding AI's Intelligence Comparison to Humans

  • Poll Question: Can AI be smarter than humans?

    • Possible responses: Yes, No, Somewhat.

Capabilities and Limitations of AI

  • AI Efficiency:

    • Excels in identifying patterns in large datasets (e.g., credit card fraud detection).

    • Recognizes visual images (e.g., for self-driving cars).

    • Operates continuously, without fatigue (e.g., AI chatbots for customer service).

  • Human Advantages:

    • Greater capability in creativity and critical thinking.

    • Understanding of emotional context and relationships.

    • Superior leadership skills and ability to make complex decisions.

AI as a Superpower for Business

  • Key Business Considerations:

    • Strategic Focus Areas:

    • Price, perceived value, cost of service, market share, and profit margin should guide AI usage.

    • Positive Aspects of AI:

    • Enhances business abilities rather than replaces human skills.

    • Aids smarter, more profitable business decisions whilst maintaining quality at reduced costs.

AI-Powered Business Decision Making

  • Scenario: Ownership of a business facing declining sales.

    • Decision Options:

    • Option 1: Guess reasons for declining sales.

    • Option 2: Utilize AI and analytics to assess customer trends and guide informed decisions.

    • Potential Outcomes:

    • Following TikTok trends for advertisement campaigns.

    • Using analytics to predict seasonal sales fluctuations and implementing targeted discounts.

    • Takeaway: Effective use of AI and analytics focuses on smarter business strategies rather than just technology use.

Introduction to Generative AI and Its Functioning

  • Definition of GenAI:

    • Large Language Models (LLMs) as foundation models (e.g., GPT-4, BART, MPT-7B) focused on language processing.

    • Functionality: LLMs are designed to understand and generate human language by leveraging deep learning neural networks trained on extensive datasets.

    • Generative AI: Sub-field aiming to produce new content across various formats (text, images, audio, etc.).

  • How Chat Models Work (broad overview):

    1. Pre-training:

    • Learning general language patterns through extensive text datasets and next-token prediction processes.

    1. Instruction Tuning:

    • Learning to follow specific instructions (e.g., summarization, classification).

    1. Alignment:

    • Training outputs to ensure they are helpful and safe through human feedback and guidelines.

Text Generation Mechanics of LLMs

  • Text Generation Theory:

    • LLMs create text through repeated next-token prediction.

    • Important implication: outputs can be unpredictable and non-factual.

  • Illustration of Token Prediction:

    • Example:

    • Input: "My favorite drink is" → Output: "bubble".

    • Input: "My favorite drink is bubble" → Output: "tea".

    • Input: "My favorite drink is bubble tea" → Output: "!".

    • Context Dependency:

    • Outputs can significantly vary due to slight input changes.

Distinction Between Analytics, AI, and GenAI

  • Analytics:

    • Framework: Data → Insight → Human Decision.

    • Tools involve dashboards, KPIs, scenario tables focusing on understanding what happened, why it happened, and implications of what could occur.

  • AI/ML:

    • Framework: Learning patterns from examples (A → B).

    • Provides outputs like scores, forecasts, classifications, and recommendations based on new input data.

  • GenAI:

    • Framework: Content generation focused on producing new pages, images, and code.

    • Most effective in drafting and acting as a thinking partner, not as a definitive authority.

    • Tip: Mastery of expressing tasks as A → B indicates understanding of AI's application.

Application of GenAI Across Business Areas

  • Use Case Analysis:

    • AI/ML often provide predictive recommendations, while GenAI is specifically for drafting deliverables.In practice, systems frequently harness both AI/ML and GenAI.

  • Business Domains and Examples:

    • Marketing:

    • A: Customer behavior, campaign exposure → B: Dinner conversion probability.

    • Task: Draft campaign concepts + KPI table.

    • Finance:

    • A: Economic drivers + seasonal trends → B: Cash-flow forecast.

    • Task: Audit memo + variance story drafting.

    • Accounting:

    • A: Invoice details → B: Fraud alerts.

    • Task: Draft audit memo and identify exceptions.

    • Operations:

    • A: Order details → B: Delay forecast.

    • Task: Standard Operating Procedure (SOP) draft + escalation checklist.

    • Human Resources:

    • A: Employee signals → B: Attrition risk evaluation.

    • Task: Formulate interview questions based on required competencies.

  • Quiz Preparation:

    • Emphasis on classifying different tasks (Analytics, AI, GenAI).

Capabilities of LLMs in Business Contexts

  • Three Task Families of LLMs:

    • C - Create: Effective in producing first drafts, options, and summaries.

    • Example Prompt: "Suggest three names for a new student-run analytics newsletter."

    • Example Output: 1) InsightPulse 2) DataBrief 3) Signal & Story.

    • Example Task: “Rewrite this update to an executive,” yielding concise action items including decision and impact details.

    • U - Understand: Proficient in tagging, routing, and basic classification tasks.

    • Example Text: "My order arrived damaged and I need a replacement ASAP."

    • Extracted Sentiment: Negative. Categorized as Replacement request with high priority.

    • Additional Utility: Extracting fields from invoices, summarizing meeting notes, and identifying risks in policies.

    • G - Guide: Adept as conversational agents and workflow guides by clarifying needs and providing structured checklists.

    • Example Interaction: User requests a dashboard for an event, AI queries decisions it should support, target audience, KPIs, and data sources.

    • Professional Caveat: Reinforces the notion that LLMs generate plausible, contextually appropriate text but should always be verified, especially in high-stakes scenarios.

Overview of LLM Benefits in Business

  • LLMs as Flexible Text Engines: LLMs can draft, transform, extract, and categorize text which can result in business-ready content.

  • Task Families and Business Applications:

    • Create/Draft: Generates initial content drafts, targeting business documents such as emails, briefs, reports, etc.

    • Summarize/Synthesize: Condenses lengthy narratives, pulls out key insights, and analyzes comparisons.

    • Extract/Structure: Streamlines information into structured tables or fields from disorganized text.

    • Classify/Label: Assigns evaluative tags such as sentiment and risk levels to different texts.

    • Transform: Rewrites material for appropriate tone and audience, translates text, and simplifies jargon into layman’s terms.

    • Pro Tip: View LLM outputs as preliminary drafts necessitating further contextual refinement and elevation to produce executive-ready materials.

Personal Experience with GenAI Tools

  • User Reflection:

    • Assessment of how GenAI tools have been utilized across various aspects (Classwork, Work/internships, Personal life).

    • Question posed: How frequently do you use GenAI applications?

Shifting Paradigms in Information Seeking with GenAI

  • Role of GenAI:

    • Acting as a thought partner in information searching and decision-making processes.

  • Source verification is crucial: Highlighting the importance of context in information retrieval by directing towards credible sources (links to Britannica).

Strategies for Employing GenAI Output for Business Applications

  • Formula for Effective GenAI output:

    • Structure: Context, Evidence, and Decisions transform GenAI output into legitimate business deliverables.

    • Components:

    1. Context: Clearly outline goals, audience characteristics, constraints, and any data you possess.

    2. Evidence: Support with metrics, dashboards, and authoritative documents plus swift verifications.

    3. Decision: Clearly articulating recommendations/directions including responsible parties and subsequent steps.

    • Professional Standard: Stating merely “AI said so” is an inadequate justification for actions taken.

Trusting GenAI Outputs and Boundaries of Application

  • Confident Use: Appropriate for tasks such as drafting, brainstorming, rewriting, summarizing, and first-step analysis.

  • Cautious Use: Highly sensitive tasks such as financial reporting, legal matters, and compliance should entail thorough scrutiny before decisions are taken.

Task Delegation Consideration for GenAI Today

  • Choices for Delegation:

    • Examples of tasks suitable for GenAI delegation (with requisite review):

    • Rewrite an email for clarity.

    • Draft a concise 1-page executive summary.

    • Generate innovative campaign concepts.

    • Create checklists.

    • Approve financial reports (caution advised).

The Nature of GenAI and Misconceptions

  • Strengths of GenAI:

    • Exceptional for tasks such as drafting emails, reports, and SOPs, summarizing information, generating creative options, and preliminary classification.

  • Notable Limitations:

    • GenAI should not be perceived as a truth generator, nor a replacement for customer judgment, nor a secure vault for sensitive data.

    • It is not the substitute for computational accuracy necessary in business operations.

  • Mental Framework:

    • Position GenAI as a supportive assistant in business contexts, where the decision-making responsibility lies with the individual.