|
| 1 | +# Draft: Lecture Conversion (Weeks 6, 7, 9, 10) |
| 2 | + |
| 3 | +## Requirements (confirmed) |
| 4 | +- Convert 10 lectures (L18-L27) from old Marp style to "box-first" gold standard matching L17 |
| 5 | +- Old style uses: `paginate: true`, `header:`, `footer:`, `<!-- _class: lead -->`, `<div class="columns">`, `<div class="callout info/tip/warning">`, emojis in titles, Title Case |
| 6 | +- New style uses: exact frontmatter (no paginate/header/footer), box divs (definition-box, note-box, tip-box, example-box, warning-box, important-box), sentence case, no emojis in titles, no multi-column layouts |
| 7 | +- Every lecture: title slide + learning objectives + ... content ... + Questions? slide (exact format) |
| 8 | +- L24 is SPECIAL: currently has RAG content but should be "Agents and Tool Use" — needs NEW CONTENT, not just restyling |
| 9 | +- Batch in waves of 3-4 for quality verification between waves |
| 10 | +- Final step: compile all + commit |
| 11 | + |
| 12 | +## Technical Decisions |
| 13 | +- Compile command: `../template_deck/compile.sh lectureN.md` from week directory |
| 14 | +- Output format: both HTML and PDF (default) |
| 15 | +- No test infrastructure needed — verification is visual (compile success + slide review) |
| 16 | + |
| 17 | +## Lecture Inventory |
| 18 | + |
| 19 | +| Lecture | File | Lines | Week | Topic | Special Notes | |
| 20 | +|---------|------|-------|------|-------|---------------| |
| 21 | +| L18 | slides/week6/lecture18.md | 729 | 6 | BERT Deep Dive | Standard conversion | |
| 22 | +| L19 | slides/week6/lecture19.md | 699 | 6 | BERT Variants | Standard conversion | |
| 23 | +| L20 | slides/week6/lecture20.md | 815 | 6 | Applications of Encoder Models | Standard conversion | |
| 24 | +| L21 | slides/week7/lecture21.md | 776 | 7 | GPT Architecture | Standard conversion | |
| 25 | +| L22 | slides/week7/lecture22.md | 954 | 7 | Scaling Up to GPT-3+ | Standard conversion | |
| 26 | +| L23 | slides/week7/lecture23.md | 996 | 7 | Implementing GPT from Scratch | Heavy code, all in example-boxes | |
| 27 | +| L24 | slides/week9/lecture24.md | 903 | 9 | Agents and Tool Use (currently RAG!) | CONTENT REWRITE needed | |
| 28 | +| L25 | slides/week9/lecture25.md | 836 | 9 | MoE & Efficiency | Standard conversion | |
| 29 | +| L26 | slides/week9/lecture26.md | 921 | 9 | Ethics, Bias, Safety | Standard conversion | |
| 30 | +| L27 | slides/week10/lecture27.md | 432 | 10 | Final Project Work Session | Short, simpler | |
| 31 | + |
| 32 | +## "Up next..." Teasers (confirmed from course schedule) |
| 33 | + |
| 34 | +| Lecture | Teaser for "Up next..." | |
| 35 | +|---------|------------------------| |
| 36 | +| L18 | "BERT variants — RoBERTa, DistilBERT, ALBERT, and how to make BERT better!" | |
| 37 | +| L19 | "Applications of encoder models — real-world NLP with BERT and friends!" | |
| 38 | +| L20 | "GPT architecture — autoregressive generation and the decoder-only transformer!" | |
| 39 | +| L21 | "Scaling up — from GPT-2 to GPT-3, scaling laws, and the rise of ChatGPT!" | |
| 40 | +| L22 | "Implementing GPT from scratch — building a language model step by step!" | |
| 41 | +| L23 | "Week 8 is off — use it to work on your final project! See you in Week 9." | |
| 42 | +| L24 | "Mixture of Experts and efficiency — how to make giant models practical!" | |
| 43 | +| L25 | "Ethics, bias, and safety — the responsible side of language models!" | |
| 44 | +| L26 | "Final project presentations — get ready to show off your work!" | |
| 45 | +| L27 | N/A (last lecture — teaser: "Good luck with your presentations!") | |
| 46 | + |
| 47 | +## Wave Structure (user preference: batches of 3-4) |
| 48 | +- Wave 1: L18, L19, L20 (Week 6 — all BERT, straightforward conversions) |
| 49 | +- Wave 2: L21, L22, L23 (Week 7 — GPT, L23 is code-heavy) |
| 50 | +- Wave 3: L24, L25, L26 (Week 9 — L24 needs new content) |
| 51 | +- Wave 4: L27 + compile all + commit (Week 10 + finalization) |
| 52 | + |
| 53 | +## Final Decisions (from user answers) |
| 54 | + |
| 55 | +### L24 Content Strategy: WRITE FROM SCRATCH |
| 56 | +- Create entirely new content on agents/tool use |
| 57 | +- Topics: function calling, ReAct framework (Yao et al. 2023), Toolformer (Schick et al. 2023), autonomous agents |
| 58 | +- Discard existing RAG content (redundant with L17) |
| 59 | + |
| 60 | +### L27 Teaser: REFLECTIVE CLOSING |
| 61 | +- "That's a wrap! From ELIZA to autonomous agents — what a journey. Good luck on your final projects!" |
| 62 | + |
| 63 | +### Announcements: INCLUDE WHERE APPLICABLE |
| 64 | +- L21: Assignment 4 Due (Feb 16), Final Project Released (Due Mar 9), Assignment 5 Available (Optional/Extra Credit) |
| 65 | +- L23: Week 8 off (no classes Feb 23-27) |
| 66 | +- L27: Final Project Due (Mar 9, 11:59 PM EST) |
| 67 | + |
| 68 | +### Manim Animations |
| 69 | +- Use same approach as L15: generate GIF with manim, process with `gifsicle --no-loopcount`, embed as `` |
| 70 | +- Infrastructure exists in `slides/week5/animations/` |
| 71 | +- Use for complex equation visualizations where appropriate |
| 72 | + |
| 73 | +## Open Questions |
| 74 | +- None remaining — all decisions confirmed |
| 75 | + |
| 76 | +## Scope Boundaries |
| 77 | +- INCLUDE: Convert all 10 lectures to box-first style, compile, commit |
| 78 | +- EXCLUDE: Modifying any already-converted lectures (L1-L17), slides/README.md updates, CI changes |
0 commit comments