Lecture 1 - Welcome to CDS593!

Understanding how LLMs actually work (transformers, attention, the "magic")
Building things: RAG systems, agents, applying concepts to real projects
Preparing for industry and understanding a technology that's reshaping the world
Some of you: approaching AI critically, wanting to understand before forming opinions

Who are YOU?

You're excited about:

Understanding how LLMs actually work (transformers, attention, the "magic")
Building things: RAG systems, agents, applying concepts to real projects
Preparing for industry and understanding a technology that's reshaping the world
Some of you: approaching AI critically, wanting to understand before forming opinions

You're a little nervous about:

PyTorch (several of you have never used it - that's okay!)
Git (it gets easier the more you use it)
Keeping up with the material / time management
The two midterms (we'll do lots of practice and review)

Who are YOU?

You bring a range of backgrounds:

Some of you have built LLM-based systems and co-authored ML papers
Some of you haven't taken a deep learning course yet
This course is designed for all of you

My hope: you'll learn a lot from each other.

I may intentionally mix groups based on background to facilitate peer learning.

Who are YOU?

You're good at things I'm going to lean on:

Resilience and persistence through difficult material
Public speaking and explaining ideas to others
Writing (professional and creative)
Theory and math
Creating visualizations and clear documentation
Bringing people together around a project
Asking questions and questioning others' thinking

A note on names

I want to learn all your names - please be patient with me for the first couple weeks!

If I mispronounce your name, please correct me. I'd rather be corrected than keep getting it wrong.

When you're here, you're HERE

We'll have discussions and activities every class
Laptops away unless we're actively using them
I might cold-call (gently!)
If you're too busy to engage, that's okay - but please don't come to class

About participation (10% of your grade)

You can engage in different ways - pick 2-3 that work for you:

Participation in lecture
Discussion section or office hours attendance
Contributing on Piazza (answering peers' questions, sharing resources)
Peer help and feedback

Twice this semester you'll write a short self-assessment making a case for your participation grade. I'll review and confirm or adjust.

Turning to the content with an Ice Breaker

Question: What's one thing you hope AI can do in the future?

What problems could AI solve? What would make your life easier? What would just be cool?

2-3 follow-ups

What even IS a Large Language Model?

A neural network trained on massive amounts of text to predict the next token (word/piece of word) that somehow develops remarkable abilities to understand, reason, and generate language

A (Very) Brief History

Natural Language Processing (NLP) has been around since the 1950s

Goal: Make computers understand and generate human language

1950s: Alan Turing's "Computing Machinery and Intelligence" (1950) (the Turing Test)
1954: Georgetown-IBM experiment - first machine translation (Russian to English)
Early approaches: hand-coded rules, symbolic AI
Why it was hard: ambiguity, context-dependence, world knowledge

The Journey to LLMs

1950s-1990s: Rule-based systems

1990s-2000s: Statistical methods (bag-of-words, n-grams)

2013: Word embeddings (Word2Vec) - words become vectors!

2014-2017: RNNs and LSTMs for sequence modeling

2017: Transformers - "Attention is All You Need"

The Transformer Revolution (2017-Present)

2018: BERT (Google) - bidirectional understanding

2018: GPT-1 (OpenAI) - 117M parameters

2019: GPT-2 (OpenAI) - 1.5B parameters - "too dangerous to release"

2020: GPT-3 (OpenAI) - 175B parameters - few-shot learning!

2022: ChatGPT launches - AI goes mainstream

2023: GPT-4, Claude 2, LLaMA 2, Gemini - the race is on

2024-2025: Agents, reasoning models (o1), Claude Sonnet 4

The Pace of Change

Image generation - results

Image generation - policy

Code generation

From fancy autocomplete to building entire apps

Multimodal

Text to vision, audio, video

Context windows

4k tokens to 200k+ tokens

This course will teach you fundamentals that persist despite rapid change, and the skills to keep up with the changing landscape!

What this course is about

By the end of this course, you will:

Understand how LLMs work (not just how to use them)
Build transformers from scratch
Apply LLMs to real problems (fine-tuning, prompting, RAG, agents)
Think critically about bias, safety, and responsible deployment
Build a professional portfolio of LLM projects

For detailed topics list and schedule, see our syllabus and the website.

Ethical Questions We'll Wrestle With

This technology raises questions we don't have answers to yet:

Environmental impact - Training costs enormous energy
Psychological safety - Reports of suicidality and psychosis in some users
Bots and fakes - Proliferation of synthetic content
Impact on learning - More classes are cancelling graded homework
Artist and author rights - Unpaid labor used to train models
Future of knowledge - What happens to deep expertise and persistence?
The big questions - AI consciousness? Existential risk?

We won't solve these, but we'll think carefully about them throughout the semester.

Course Website Tour & Syllabus

Let's look at the course website

You're already here! Take a moment to explore:

Syllabus
Course Schedule
Lecture notes (this page!)

What you'll find on the website

Full syllabus with course policies
Weekly schedule with due dates
Lecture notes for every class
Links to resources

Bookmark this page - it's your home base for the semester

How this course works

No traditional homework! Instead:

Weekly reflections (200-500 words)
Lab notebooks (hands-on experimentation)
2 portfolio pieces (polished projects)
2 midterm exams (theory, no AI)
Final project (build something cool!)

All work goes in your GitHub portfolio - you'll have something to show employers!

Compute Resources

Towards the end of the course (and for your final project), you'll need more compute than your laptop can provide.

Recommended approach: Google Colab with education credits

Alternative: BU's Shared Computing Cluster (SCC)

If you find you need more compute than that, talk to us.

For first discussion (Tuesday): Try to have GitHub and a Colab account set up. Bhoomika can help troubleshoot any issues.

A note on how I teach

There will be times when I think I can explain something to you most effectively in person.

And there will be times when I think your best opportunity to learn comes from a YouTube video, a blog post, or other resources.

I'll be intentional about which is which. When I assign prework, it's because I genuinely think that's the best way for you to learn that material - not because I'm offloading teaching.

Key Course Policies

A few highlights before we dive into the full syllabus:

AI use for coding: Encouraged! Use it as much as you want. (Correspondingly: high expectations for project quality)

AI use for reflections: Please write in your own voice, no AI

Exams: No notes - just you and the concepts

Late work: 100% on time, 90% one day late, 80% two days late, exceptions are rare

Struggling? Reach out early! Extensions available, wellness matters

Syllabus Activity (20 min)

Time to dig into the details!

Instructions:

Form groups of 2-3 people
Grab a printed syllabus and worksheet
Work together to answer the questions
We'll reconvene in 15 minutes to discuss

Let's debrief

Essential Shell & Git Skills

What's your experience level with shell and git?

Drop hands polling

Why shell and git?

Essential skills for developers and researchers that enable efficient iteration and collaboration
Even MORE essential if you're handing the reins to AI development tools
We'll use these throughout the course - your investment now will pay off later

The command line is your text-based interface to your computer

Essential commands:

pwd                   # Print working directory (where am I?)
ls                    # List files
ls -la                # List all files including hidden ones
cd folder_name        # Change directory
cd ..                 # Go up one level
cd ~                  # Go to home directory

If you're on windows, you can use git-bash for linux-compatible command line, or learn somewhat different commands for a shell like powershell

Tips:

Use Tab for auto-completion
Use Up Arrow to repeat previous commands
Ctrl+C to cancel/abort

Shell Basics: File Operations

mkdir project_name       # Create a directory
touch filename.txt       # Create an empty file
echo "text" > file.txt   # Write text to file
cat filename.txt         # Display file contents

cp file.txt backup.txt   # Copy a file
mv old.txt new.txt       # Rename/move a file
rm filename.txt          # Delete a file

For Lab 0, you'll mostly use:

cd to navigate to your projects folder
mkdir to create your course repo folder
git commands (next slide!)

Git & GitHub Essentials

Git = version control system (tracks changes to your code)

GitHub = hosting service for git repositories (plus collaboration tools)

You'll use GitHub Classroom for this course

Git Workflow for This Course

# One-time setup
git config --global user.name "Your Name"
git config --global user.email "your.email@bu.edu"

# For each lab/assignment
git clone [repo-url]           # Get the repo from GitHub
cd repo-name                   # Navigate into it

# Work on your code, then...
git add .                      # Stage all changes
git commit -m "Descriptive message"  # Save a snapshot
git push                       # Upload to GitHub

That's it! For this course, you mostly just need: clone, add, commit, push

Git Cheat Sheet

Common commands:

git status              # What's changed?
git add filename        # Stage specific file
git add .              # Stage everything
git commit -m "msg"    # Save a snapshot
git push               # Upload to GitHub
git pull               # Download from GitHub
git log                # See commit history

Good commit messages:

"Add spam detection implementation"
"Fix typo in reflection"
"Complete Lab 1 embeddings exploration"

Pro tip: If you need to use "and" in your commit message, you're probably committing too many changes at once!

Resources for Shell & Git

For Lab 0: You just need the basics - we'll practice more as the semester goes on

Challenge the AI!

Time to see what LLMs can (and can't) do

Let's put ChatGPT and Claude to the test!

Your mission: Come up with questions or tasks that might trip them up

A few starter ideas...

Ask it to count the number of times the letter 'r' appears in "strawberry"
Ask it about very recent events (knowledge cutoff!)
Ask it to do complex multi-step reasoning
Ask it something that requires true understanding vs pattern matching
Try to get it to contradict itself

5 minutes: Pair up and try to stump the AI on your laptops

Complete the intro survey (linked on Piazza)
Set up: GitHub account, Python environment, Jupyter notebooks
Create your course GitHub repository (link to come)
Write your first reflection (see website)
Lab 0 (see website)

Coming up

Monday: AI-assisted development + Classical NLP introduction

How to use AI coding tools effectively
Bag-of-words and TF-IDF
Start of Lab 1

See you Monday!

CDS593 Syllabus Review Worksheet

Group members:

Concrete questions:

How are weekly reflections and lab notebooks submitted?
What happens if you submit work a day late?
Is attendance in discussions required?
If you get stuck on an assignment and your friend explains how to do it, what should you do?
If you have accommodations for exams, how soon should you request them?
Is there a final exam for the course?
Can you use AI tools when working on portfolio pieces?
Can you use AI tools to help write your reflections?

Open-ended questions:

What parts of the course policies seem standard and what parts seem unique?

Standard: Unique:
Identify 2-3 things in the syllabus that concern you
What strategies could you use to address these concerns?
Identify 2-3 things on the syllabus that you're glad to see
List 2-3 questions you have about the course that aren't answered in the syllabus
What kind of engagement do you think you'll focus on for participation credit?

Lauren's CDS593 Materials