CMU Launches '10-202: Introduction to Modern AI' - A Hands-On Course Building Chatbots from Scratch

Carnegie Mellon University is launching a new course this spring that takes a radically different approach to teaching artificial intelligence. Instead of traditional theory-heavy lectures, 10-202: Introduction to Modern AI will guide students through building their own AI chatbot from scratch.

The course, taught by Professor Zico Kolter, focuses specifically on the machine learning methods and large language models (LLMs) that power systems like ChatGPT, Gemini, and Claude. Rather than covering the broad academic field of AI, it zeros in on the practical techniques that make modern chatbots work.

Here's what makes this course unique: students will implement a basic LLM in just a few hundred lines of code. The curriculum walks through supervised machine learning, neural networks, transformers, and the self-attention mechanisms that make chatbots possible. By the end, students will have written code that runs an open-source LLM and trained it on their own data.

Course Structure and Requirements

The course meets Mondays, Wednesdays, and Fridays from 9:30-10:50 AM in Tepper 1403, with Friday sessions reserved for review or makeup lectures. Students need basic Python programming skills (equivalent to 15-112 or 15-122) and some calculus background. Familiarity with linear algebra and probability helps but isn't required.

Grading breaks down into 20% homework assignments, 40% homework quizzes, and 40% exams (two midterms and a final). The homework component is particularly hands-on - students develop a minimal AI chatbot through seven programming assignments using Colab notebooks or Marimo notebooks.

The Learning Journey

Assignments progress logically from fundamentals to advanced topics:

• Homework 0 covers autograding and programming basics
• Homework 1 introduces linear algebra and PyTorch
• Homework 2 teaches automatic differentiation and training linear models
• Homework 3 focuses on neural network implementation
• Homework 4 implements transformers
• Homework 5 builds a minimal LLM
• Homework 6 covers supervised fine-tuning and chat training
• Homework 7 explores reinforcement learning

Each assignment includes a 15-minute in-class quiz testing both conceptual understanding and the ability to replicate key code patterns.

AI Policy - Practice What You Teach

In an interesting twist, the course allows students to use AI assistants for homework and programming assignments, recognizing that learning to use these tools is itself a valuable skill. However, students are strongly encouraged to complete their final submissions without AI assistance. The rationale is straightforward: while AI can be helpful for learning and implementation, over-reliance may hinder deep understanding of the material.

During in-class evaluations - both homework quizzes and exams - no AI tools or external materials are permitted. This policy reflects the course's philosophy that students need to understand the underlying mechanics, not just how to prompt an AI system.

Online Offering

A free online version runs simultaneously with the CMU course, starting January 26 with a two-week delay. This makes all lecture videos, assignments, and autograded materials accessible to anyone worldwide. However, the online version doesn't include quizzes, midterms, or finals, and lacks the in-person TA support and office hours available to CMU students.

The course represents a practical, hands-on approach to AI education that mirrors how these systems are actually built in industry. Rather than abstract theory, students gain experience with the specific architectures and training techniques that power today's most popular AI applications.

For students interested in understanding not just how to use AI systems but how to build them, 10-202 offers a comprehensive introduction to modern AI implementation. The course materials, including lecture videos and assignments, will be available online two weeks after the in-person offering, making this cutting-edge AI education accessible to a global audience.