| Week | Date | Content | Exercise | Notes |
|---|---|---|---|---|
| 1 | 9/7 | Overview of the course | HW1 released: Sharing your experiences of using (and NOT using) LLMs | |
| 2 | 9/14 | New Research Mindsets and the Foundations of LLMs 1. Research mindsets in the Gen AI era: Vision, Ideas, and Execution; digitizing and structuring data; how to ask questions and develop research problems 2. Why understanding the principles helps you use the tools better 3. The Transformer architecture and positional encoding 4. Model training pipeline: Pretraining, Finetuning, RLHF, RLVR | ||
| 3 | 9/21 | Scaling Up and Finetuning; Limitations and Hallucinations of LLMs 1. Sparse architectures: Mixture of Experts (MoE) 2. Parameter-efficient finetuning (PEFT): LoRA, DoRA, CeRA 3. Limitations of the context window 4. Why hallucinations occur and why they cannot be fully eliminated 5. Model biases and case studies | HW1 due HW2 released: AI-assisted programming and system implementation | |
| 4 | 9/28 | Holiday | ||
| 5 | 10/5 | Asynchronous online teaching | Materials: 1. Towards end-to-end automation of AI research 2. The Agentic Researcher: A Practical Guide to AI-Assisted Research in Mathematics and Machine Learning 3. The AI Scientist: Towards Fully Automated Open-Ended Scientific Discovery | |
| 6 | 10/12 | From Programming to Vibe Coding 1. From writing code to describing architectures in natural language 2. Comparison of Copilot, Claude Code, and Yolo mode 3. Building a minimum viable prototype (MVP) with coding agents | ||
| 7 | 10/19 | Principles of Agentic AI 1. Core components: Planning, Memory, and Tool-use 2. The ReAct framework 3. Building tools for AI: API integration and function calling 4. Case study: Karpathy's autoresearch and its limitations | ||
| 8 | 10/26 | Holiday | ||
| 9 | 11/2 | Guest lecture - 鍾筑安 (Judy) | HW2 due (source code and AI collaboration log) HW3 released: AI-assisted mini-thesis (final research topic proposal) | |
| 10 | 11/9 | Software Engineering Practices and Harness Engineering 1. Version control with Git and CI/CD 2. Writing tests and code review 3. The importance of scoping and automated testing in the agentic AI era | ||
| 12 | 11/16 | Choosing Research Topics and MVP Validation 1. How to assess the value of a research topic 2. Literature review and gap analysis with agentic AI 3. MVP: Minimum Viable Paper and Minimum Viable Product | ||
| 11 | 11/23 | Guest lecture - 許煜松 (Kuma) | ||
| 11 | 11/30 | Guest lecture (Speaker: TBA) | ||
| 14 | 12/7 | Guest lecture (Speaker: TBA) | ||
| 15 | 12/14 | Asynchronous online teaching Thesis Writing and Academic Ethics1. Academic English writing skills and judgment 2. Structured typesetting with LaTeX 3. Ideating, experimenting, and writing in parallel 4. Disclosing the scope of AI contributions and academic accountability | ||
| 16 | 12/21 | Final Mini-Thesis Presentations 1. Research results, code logic walkthrough, and AI collaboration log 2. Q&A and course review | HW3 due (mini-thesis PDF, source code, and collaboration log) |