Overview

Description

CALENDAR DESCRIPTION:

Special topics in computing science which are of current interest or are not covered in the regular curriculum will be offered from time to time depending on availability of faculty and on student interest.

COURSE DETAILS:

CMPT 218 introduces foundation models (large pre-trained models), with a focus on generative AI and AI agents as tools for academic, professional, and personal workflows. Students learn how modern AI systems work and how to use them effectively through both a chat interface and API access. The course explains topics such as prompting, sampling parameters, retrieval-augmented generation, and tool calling, and how these affect outputs. Students also learn how to build workflows and analyze data with AI, assess risks, and communicate responsible use. Topics include the capabilities and limitations of AI models, prompting, APIs, agents, privacy, and responsible use.

The course is open to students from any faculty and assumes no prior programming or AI background. It is especially recommended for students who may need to use AI in academic or professional contexts (e.g., for data analysis). Activities include no-code tools and guided Colab notebooks. A typical week includes an interactive lecture, a guided activity, and a discussion. Activities and homework can be completed on personal laptops.

Note on difficulty: Although the course requires no coding, it is NOT a surface-level overview. The students will be scaffolded and supported at each step so they genuinely learn to apply AI to real tasks. 

Note on credit: This course is intended for students outside the CMPT AI/ML stream. Students who have completed, or are concurrently enrolled in, CMPT 310, CMPT 311, or any 400-level CMPT course should NOT take it for credit.

Topics (content may be adjusted based on new developments in AI systems):

1. Foundation models: What are foundation models? How does a language model differ from a chatbot? 
2. Data & training: data, scale, tokens, embeddings, and training at a conceptual level.
3. Inference: prompting, interaction, system constraints, prompt sensitivity.
4. Issues: hallucination, verification, citations, uncertainty, evidence, and AI-detection.
5. Augmenting models: retrieval-augmented generation, search, and using external sources responsibly.
6. Evaluation: benchmarks, human evaluation, failure analysis, claims auditing.
7. Multimodal models: images, speech, video, documents.
8. Fairness: bias, representation, labour, privacy, copyright, and environmental costs.
9. Reasoning models: what "reasoning" means; accuracy, cost, implications.
10. AI agents: goals, planning, tool use, permissions, autonomy levels, logging, agent-trace audits.
11. Governance and policy: disclosure, procurement, risk tiers, accountability, oversight.

The class will also include a guest lecture and final project presentation.

COURSE-LEVEL EDUCATIONAL GOALS:

By the end of the course, successful students will be able to:

1. Explain the basic ideas behind foundation models, including tokens, training data, embeddings, prompting, retrieval, fine-tuning, and model evaluation.
2. Make informed decisions on whether, when, and how to apply AI to their workflows.
3. Use generative AI tools with explicit goals, constraints, verification steps, and disclosure.
4. Identify common failure modes: hallucination, bias, privacy leakage, prompt sensitivity, brittle evaluation, and inappropriate automation.
5. Compare AI systems or workflows using evidence rather than vendor claims.
6. Analyze the costs and benefits of using AI in different settings.
7. Build a small AI-assisted workflow (e.g., data processing) using plug-and-play code examples.

Grading

Materials

MATERIALS + SUPPLIES:

No textbook is required. Additional reading and watching materials will be provided for each class as needed.

RECOMMENDED READING: