Saturday, April 4, 2026

2.1 Overview of AI and ML

 

2.1 Gen AI: AI Literacy

Sat, 04 Apr 26

1 Overview

  • Session structured in two parts:

    • Wrap-up of Python refresher (AI-powered coding tools)

    • Introduction to AI Literacy

  • Engage and Think prompt: CEO of mid-size e-commerce company exploring AI — which business functions to prioritise?

    • Options: machine learning for customer behaviour analysis, deep learning for fraud detection, or AI-powered automation for marketing and logistics

  • Learning objectives:

    • Describe fundamental concepts of AI

    • Differentiate between machine learning and deep learning

    • Apply knowledge of neural networks and deep learning models to real-world applications

Intro to AI — What is AI?

  • Branch of computer science focused on creating autonomous systems that mimic human intelligence

    • Tasks include: reasoning, learning, problem solving, perception

  • AI is the broader field; subcomponents nest within it:

    • AI > Machine Learning > Deep Learning

Evolution of AI

  • 1950s–1980s: Rule-based AI (if-then logic)

  • 1990s–2010s: Machine learning era — learn from data, statistical models

  • 2010s–present: Deep learning — neural networks, self-learning

  • 2020s–present: Generative AI revolution — AI creates text, images, and code

  • Key reason for recent acceleration: infrastructure finally caught up with ideas

    • Graphics Processing Units (GPUs) and Tensor Processing Units (TPUs) enabled large-scale computation

    • Cloud computing removed need for private data centres

    • Deep Seek (China) trained a powerful model at a fraction of the cost of US counterparts — caused Nvidia stock to drop temporarily

  • AI Winters (1970s): period of near-zero funding and research activity

    • Ideas existed but infrastructure did not yet support them

Types of AI

  • Narrow AI (Weak AI)

    • Designed for specific tasks

    • Examples: chatbots, recommendation systems

  • General AI (Strong AI)

    • Hypothetical AI with human-like intelligence and reasoning

    • Not yet achieved

  • Super Intelligent AI

    • Theoretical AI that surpasses human intelligence

    • The “singularity” — concern that AI could reverse-control humans

    • Not currently achieved; remains a subject of debate and speculation

Key Components of AI

  • Data, algorithms/models, computing power

  • Natural language processing

  • Computer vision

  • Robotics and automation

  • Edge AI and IoT integration

    • Edge devices: phones, factory robots, IoT sensors, weather monitors

    • Can operate autonomously without central connectivity

  • Ethics and responsible AI

  • AI frameworks and tools

Real-World Applications

  • Virtual assistants and chatbots

  • Recommendation systems (e-commerce, social media feeds, YouTube)

  • Smart home devices and IoT

  • Autonomous vehicles (Waymo, Tesla)

  • Healthcare and medical

    • Cancer diagnosis, protein synthesis, genetics

    • Human Genome Project now achievable far faster with AI

  • Finance and fraud detection

  • AI in education

    • Khan Academy’s Khanmigo — free AI tutor, tailors content to individual learners

  • Personalisation engines (Meta, TikTok, etc.)

Benefits of AI

  • Automation and efficiency

  • Enhanced decision making

  • Personalisation

  • Improved accuracy

  • 24/7 availability

  • Cost savings

  • Enhanced security

  • Scalability and global reach

    • Internet access (e.g. via Starlink) equalises productivity regardless of location

  • Innovation and new business models

  • Agility, faster decision-making, and competitive advantage

Challenges of AI

  • Data privacy and security

    • Key enterprise concern: will data be exposed externally to providers like Google or Amazon?

  • High implementation costs

    • Powerful model tokens are expensive

    • Skilled AI engineers are scarce and costly to hire

  • Bias and fairness

    • Models trained on publicly available data inherit its biases

    • AI scales biases at speed — requires active mitigation

  • Lack of transparency

    • Emergent behaviours: models do things they weren’t designed to do

    • Mitigation: collect trajectory data to observe reasoning, tool calls, and decisions

  • Dependence on quality data

    • Retrieval-Augmented Generation (RAG) and fine-tuning require clean data

    • Noisy input = poor quality output

  • Job displacement

    • Companies replacing junior engineers with fewer senior engineers who leverage AI heavily

    • Risk: no pipeline of junior talent to replace retiring seniors

    • Instructor’s view: a talent crisis is likely in the next few years

  • Ownership and authenticity of AI-generated content

    • Responsibility lies with the entity that publishes the content

    • Regulators struggle to keep pace with private sector innovation

Rise of Machine Learning

  • Shift from rule-based systems to data-driven approaches

  • From expert systems to statistical learning

  • Advancement in algorithms

  • Neural network revival

  • Enabled by availability of GPUs for large matrix computations

AI-Powered Coding Tools (Python Refresher Wrap-Up)

  • AI coding tools: GitHub Copilot, Gemini, Amazon Q, Claude Code, Tab Nine

  • Core capabilities: generate code, optimise, refine, debug — all via natural language

  • Impact on job market:

    • Junior developer roles declining; companies hiring fewer, more senior engineers with strong AI skills

    • Senior engineers with AI skills can double productivity

    • Companies motivated by ROI and shareholder returns, not just headcount reduction

  • Code quality and AI:

    • AI-generated code is not automatically production-ready

    • Quality of output depends heavily on quality of instructions given

    • Human review still required before deploying to production

    • Example risk: AI-generated banking system code could hallucinate and misroute funds

Effective AI Prompting

  • A prompt = instruction given to an AI application to produce a desired output

  • Quality of prompt directly determines quality of output

    • Two people using the same tool (e.g. Copilot) will get very different results based on prompt clarity

  • Key prompting techniques:

    • Use precise, unambiguous instructions

    • Provide examples — show AI the expected output format

    • Define constraints — specify what AI should and should not do

    • Break down complex tasks into smaller steps (Chain of Thought)

    • Iterate and refine — don’t rely on a single prompt

  • Context window management:

    • LLMs have a finite context window (token limit)

    • As tokens are consumed, reasoning quality degrades and hallucination increases

    • Best practice: start a new session when context reaches ~50–70% capacity

  • Reflection technique:

    • Copy error or unwanted output, paste back into chat, explain what went wrong

    • Helps the model self-correct without manual back-and-forth

  • Spec-driven development:

    • Write a specification (e.g. in a .md file) before writing any code

    • Feed the spec to AI; ask it to evaluate its own output against the spec

    • Shifts the engineer’s role toward system design and code review

GitHub Copilot Setup (VS Code)

  • Requires a GitHub account (free tier sufficient)

  • Install the GitHub Copilot Chat extension from VS Code Marketplace

  • Sign in with GitHub credentials to activate

  • Configuration steps:

    • Go to Settings > Extensions > GitHub Copilot

    • Enable plain text and markdown completions (set to true)

    • Enable inline auto-completions under Copilot Editor settings

  • Multiple AI models available within Copilot Chat (GPT-4o, Claude, Gemini, etc.)

  • Can also add local models via Ollama within VS Code

  • Common issues encountered in session:

    • “Extension disabled — workspace not trusted” error

    • Missing GitHub Copilot Chat vs. older Copilot extension version

    • Python environment not configured in VS Code

Why AI Skills Are Non-Negotiable

  • AI has become integral to every sector — not optional

  • Competitive advantage: candidates with AI skills will be preferred over those without

  • Applies equally to employees and business owners

  • Staying sharp: continue verifying AI output, keep learning domain-specific knowledge

    • Value shifts from speed of execution to system thinking and review

Next Steps / Action Items

  • Michael Chang (all students)

    • Complete Demo 1: install and configure GitHub Copilot in VS Code

    • Complete Demo 2: configure Copilot settings (plain text, inline completions)

    • Try Demo 3: use Copilot to generate Python functions using inline comments and docstrings

    • Practice using Copilot for OOP (classes, methods, constructors) covered in previous sessions

    • Set up a GitHub account if not already done — required for portfolio and Copilot activation

    • Review Python refresher demos before attempting the two assessment projects and timed test (~2.5 hours)

    • Fill in the survey link posted by Rashmi in the Zoom chat

  • Students with setup issues (Sebastian, Gina, Praga, others)

    • Remain after class for screen-share troubleshooting of VS Code / Copilot configuration

  • Next session (tomorrow):

    • Start with Machine Learning overview

    • Move into Introduction to Transformers, Advanced AI Models, and Natural Language Processing
on
Labels: ,

No comments:

Post a Comment

Subscribe to: Post Comments (Atom)