How to use this list
Items marked [REQUIRED] must be completed by the listed week. Items marked [RECOMMENDED] will deepen your understanding but are not required. Items marked [OPTIONAL] are for students who want to go further or whose project aligns with that area. Use Keshav's 3-pass method: first pass = abstract + intro + conclusion; second pass = full reading; third pass = you could present it.
Videos
3Blue1Brown — Neural Networks Series (YouTube · 4 videos · ~20 min each)
Best visual explanation of how neural networks learn. Covers gradient descent, backprop, and why depth helps. Watch all 4 before Wednesday of Week 1.
Andrew Ng — AI For Everyone (Coursera · ~6 hours · free to audit)
Non-technical overview of AI in organizations. Excellent for understanding what AI can and cannot do. Good if you're not from a CS background.
Corey Schafer — Pandas Tutorial Series (YouTube · 10 videos)
The most direct pandas tutorial for this project. Covers loading CSVs, filtering, groupby, and basic analysis — exactly what you need for your experiments.
Andrej Karpathy — "The State of GPT" (Microsoft Build 2023 · ~45 min)
How GPT-style models are trained: pre-training, RLHF, instruction fine-tuning. Very accessible. Useful for Week 3 context on prompt engineering.
Andrej Karpathy — "Let's Build GPT from Scratch" (YouTube · ~2 hours)
You will genuinely understand transformers after watching this. Builds a character-level GPT from scratch in ~200 lines of PyTorch. Strongly recommended if your project involves model behavior.
Programming and Tooling
Python Crash Course — Eric Matthes (Chapters 1–6)
Variables, lists, dicts, functions, loops, files. Everything you need. Skip the pygame chapters. Available at most libraries and free as a PDF online.
CS50P — Harvard Python Course (edX · free to audit · Weeks 0–4)
High production quality. Problem sets are excellent practice. Do Weeks 0–4 if new to Python.
NLP and Transformers
Vaswani et al. — "Attention Is All You Need" (2017, NeurIPS)
The foundational transformer paper. Read: abstract, Section 1 (intro), Figure 1 (architecture diagram). Every modern LLM descends from this architecture.
First model to show pre-training then fine-tuning works dramatically better than task-specific training. Read: abstract, intro, Section 2 (model architecture).
Brown et al. — "Language Models are Few-Shot Learners" (GPT-3, 2020, NeurIPS)
Introduced few-shot prompting as a concept. Read: abstract, Section 1 (intro), Section 2 (few-shot learning definition). Conceptual foundation for all prompt engineering.
Short paper showing "think step by step" dramatically improves reasoning accuracy. Read the introduction and Figure 1. Directly applicable to your prompting experiments.
The original RAG paper. Read: abstract, intro, Section 2 (RAG model definition). Sections 3–5 if you're doing the RAG or cross-regulation project.
Research Methods
Keshav — "How to Read a Paper" (2007, ACM SIGCOMM · 3 pages)
Apply the 3-pass method to every paper on this list. Also apply it to your own write-up: would it survive someone's first-pass reading?
Heilmeier Catechism (DARPA · 1 page)
8 questions every research proposal must answer: What are you trying to do? How is it done today? What is new? Who cares? Read before writing your Week 3 project proposal.
Regulations
HHS — HIPAA Privacy Rule Summary (hhs.gov · 2 pages)
The official HHS summary of the HIPAA Privacy Rule. Read before Week 2. You don't need the full regulation — just the summary to understand the structure and key exceptions.
HHS — Permitted Uses and Disclosures (fact sheet)
Lists all ~15 permitted disclosure categories. Reference document — don't memorize it, but know where to look when a scenario triggers one of these categories.
Article 6: the six lawful bases (consent, contract, legal obligation, vital interests, public task, legitimate interest). Article 9: health data, biometric data, genetic data — the categories that overlap most with HIPAA.
CCPA — California Consumer Privacy Act (California Attorney General)
Consumer rights: right to know, right to delete, right to opt-out of sale. Read the summary page. Relevant for the cross-regulation project — "can I delete my records?" hits CCPA and GDPR simultaneously.
Financial institutions must protect customers' non-public personal information. Analogous to HIPAA but for banks and insurers. Relevant when a health insurer's billing data crosses financial and medical categories.
Requires parental consent before collecting data on children under 13. Overlaps with HIPAA and GDPR when the patient is a minor.
AI and Law
Surden — "Artificial Intelligence and Law: An Overview" (Georgia Law Review, 2019)
Accessible overview of how AI intersects with legal systems. ~30 pages but skimmable. Useful for writing your project motivation section.
GoldCoin — "GOLDCOIN: Grounding Large Language Models in Privacy Laws through Contextual Integrity" (2024)
The benchmark paper for the dataset you will work with. 137 real HIPAA enforcement scenarios with expert-validated ground truth. Read: abstract, Section 2 (dataset construction), Section 4 (results). This is the paper closest to what you are doing.
Survey of NLP applications in legal tasks: judgment prediction, legal question answering, court summarization. Good for situating your project in the broader research field.