Info 290:
Fundamentals of Generative AI

This course provides a foundation in the principles, architectures, and applications of generative artificial intelligence (GenAI). It explores how large language models (LLMs) work by guiding students through the complete model lifecycle: from core building blocks like embedding representations, attention mechanisms, and transformer architectures to adaptation techniques including in-context learning, prompt tuning, parameter-efficient fine-tuning (PEFT), and reinforcement learning (RL).

The course also explores advanced topics such as retrieval-augmented generation (RAG), agentic reasoning systems, and multimodal learning that extends beyond text to incorporate vision and other data modalities.

Programming assignments, paper readings, and a team-based final project will help students develop the technical proficiency to design, implement, and evaluate GenAI pipelines using Hugging Face, LangChain, and other open-source LLM APIs, while critically assessing their capabilities, limitations, and responsible use in real-world applications.

All materials for this course will be posted on this page.

• Fill out the course survey
• Set up Google Colab + Drive and Hugging Face

• Graduate student status
• Proficiency in Python programming, equivalent to INFO 206 A/B (including writing multi-file programs of ~200+ lines of code).

• Python 3
• JupyterLab / Jupyter Notebook workflows
• Git and GitHub (clone, commit, push)

• Build a Large Language Model (From Scratch) , Raschka S., Manning Publications, 2024 (SR)

• Submitted via Gradescope

Upon successful completion of this course, students will be able to:

• Analyze the architectural components of LLMs (embeddings, attention, Transformer blocks).
• Implement GenAI systems (fine-tuning, PEFT, RLHF) and integrate RAG + tools using Hugging Face and LangChain.
• Apply prompt engineering and in-context learning for downstream tasks without extensive retraining.
• Design end-to-end GenAI pipelines (RAG, agents, multimodal inputs).
• Evaluate GenAI systems (performance, fairness, bias, safety).
• Synthesize insights from current research literature.

We will use Ed Discussion as the primary communication channel throughout the semester.
Questions and comments related to your homework and projects (excluding code) are strongly encouraged.

Your final course grade will be calculated as follows:

Final grades will be determined by computing the weighted average of assignments, final group project, paper reading, and participation.

Grading range for this course is: A for 93 or above, A- for 90 or above, B+ for 87 or above, B for 83 or above, B- for 80 or above, C+ for 77 or above, C for 73 or above, C- for 70 and above, D+ for 67 and above, D for 63 and above, D- for 60 and above, and F for 59 and below.

Students will complete two individual assignments (each worth 15% of the final grade) designed to connect lecture concepts with data-driven, real-world applications. Late submissions are accepted up to three days past the deadline, with a 10% absolute deduction per day. Assignments will be posted on this website and submitted through Gradescope.

The final project is designed to help you develop practical skills in building generative AI pipelines for real-world applications. The maximum is 12 teams (~4 students per team).

Project grades will be determined based on the proposal, milestone, final report, and poster presentation, with adjustments made for group size and individual contributions. The final project accounts for 50% of the total course grade, distributed as follows:

• 5%: Project proposal
• 15%: Project milestone
• 20%: Final report
• 10%: Poster presentation

There is no late policy for the final project. All deliverables must be submitted on time. Detailed project guidelines will be distributed at the beginning of the semester.

Each week, 2 groups of 2 students will present selected papers in class related to GenAI concepts (max 15 minutes). Submit your slides to Gradescope and make sure to include/tag all group members in your submission.

Active engagement during class is essential for mastering GenAI concepts. Participation will be evaluated based on contributions to discussions, questions asked, and involvement in group exercises. Participation accounts for 1% per week of the total course grade.

Students may use AI chatbots (e.g., ChatGPT, Claude, Gemini, and others) to clarify concepts, debug code, generate small code snippets for study or practice, or paraphrase/revise text that they originally authored. Chatbot use must align with the principles of academic integrity and may not replace a student’s independent effort.

• Students may not submit large blocks of unmodified chatbot-generated code or use chatbots to complete entire problems.
• Students must be able to explain all submitted code and may be asked to demonstrate understanding in follow-up assessments.
• When chatbots are used, students must include a comment in their code specifying the tool and the exact prompt used (e.g., “I used ChatGPT for this problem with the prompt: ‘How to implement a bubble sort in Python?’”).
• If chatbots are used to paraphrase or revise text originally written by the student, the student must explicitly disclose how the tool was used.
• Misuse of chatbots or failure to follow this policy will result in disciplinary action as outlined in the course syllabus.

All students are expected to follow the UC Berkeley Code of Student Conduct. While the final project is a collaborative effort, all assignments must be completed independently unless explicitly stated otherwise.

All written work must be your own. If you reference the work of others, you must clearly cite the appropriate source. This applies to source code as well—if you use code from external sources (e.g., Stack Overflow, ChatGPT), you must cite its origin.

We have a zero-tolerance policy for cheating and plagiarism. Violations will be referred to the Center for Student Conduct and will likely result in failing the course.

A broad range of perspectives enhances our collective understanding. In keeping with the UC Berkeley Principles of Community, we strive to create a learning environment that respects all identities—race, gender, class, sexuality, religion, ability, veteran status, and more—and values the full diversity of thought and experience.

If you have suggestions for improving the course or if something made you uncomfortable, please contact the instructors or Student Affairs. If external factors affect your performance, do not hesitate to reach out—we are here to help.

As participants in discussions and teamwork, please honor the diversity of your classmates and maintain a respectful, inclusive dialogue.

UC Berkeley is committed to creating a learning environment that meets the needs of its diverse student body, including students with disabilities. If you anticipate or experience barriers to learning, please feel welcome to discuss your concerns with the instructor.

Students may work with the Disabled Students’ Program (DSP) to request official accommodations. More information is available at dsp.berkeley.edu.

DSP is located at 260 César Chávez Student Center. Phone: (510) 642-0518 (voice), (510) 642-6376 (TTY), Email: dsp@berkeley.edu.

Do your best to maintain a healthy lifestyle this semester by eating well, exercising, getting enough sleep, and taking time to recharge. Avoiding burnout will support your academic, professional, and personal goals.

If you feel overwhelmed, please be kind to yourself and reach out for support. Seeking help is a courageous step. Berkeley offers a wide range of emotional, physical, safety, and academic wellbeing resources, including support through the Student Learning Center and campus wellness services.