Skip to content

Add Q&A 1 slide deck + drop course page ToC#383

Draft
natolambert wants to merge 1 commit intomainfrom
qa-slide-deck
Draft

Add Q&A 1 slide deck + drop course page ToC#383
natolambert wants to merge 1 commit intomainfrom
qa-slide-deck

Conversation

@natolambert
Copy link
Copy Markdown
Owner

Summary

  • New teach/course/qa-01.md — first round of a Q&A slide deck series. Each slide is a placeholder: reader's question quoted, light context in the body, attribution in the footer. I'll talk through each one live and we can flesh out follow-up slides after.
  • Uses the title-banner layout on the title slide to make Q&A decks visually distinct from the lecture decks.
  • Drops the small Contents mini-ToC from book/templates/course.html (the page is short enough without it) and adds the Q&A deck as an inline entry in the Lectures list, after Lecture 4.

Questions in this round

Lecture 2 — IFT, Reward Models, & Rejection Sampling

  • One model vs. many for synthetic SFT data? — @mufeezamjad790 (YouTube)
  • Should teacher distribution match the base model? — @mufeezamjad790 (YouTube)
  • Unified RM benchmark across RM types on one dataset? — @jeromeeusebius (YouTube)
  • How important is balancing domain proportions in an SFT mix? — @jeromeeusebius (YouTube)

Lecture 3 — RL Motivation & Math

  • What is the shape of $\rho$? — @JGLambourne (YouTube)
  • Is GAE the only way to compute the advantage in PPO? — #382
  • Should the notation table define $T$, $K$, $G$? — #382
  • Are slides 59–61 in Lec 3 in the right order? — #382

Test plan

  • Check the rendered course page on the Pages preview: Contents nav is gone, Q&A entry shows up after Lecture 4 with PDF / Slides / Source buttons.
  • Confirm the Q&A deck builds (HTML + PDF) and the title-banner layout looks right.
  • Click through each Q slide and sanity-check the per-slide footer attribution renders.

🤖 Generated with Claude Code

Kick off a per-round Q&A series to address reader questions from the
issue tracker, Discord, and YouTube comments. First round covers
Lecture 2 (synthetic data, RM evaluation, domain balancing) and
Lecture 3 (importance sampling ratio shape, advantage estimators,
notation, slide ordering).

Also drops the Contents mini-ToC from the course page since the
section is now short enough to scan without it, and adds the Q&A
entry inline in the Lectures list.

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant