Nathan Lambert releases Q&A 2 in his free RLHF Book course

Nathan Lambert has been drip-feeding his free RLHF course alongside the book of the same name, and the latest addition is a reader Q&A. According to the course page, Q&A 2 gathers questions from lectures 5 through 7, the stretch that covered reasoning models, Direct Preference Optimization, and synthetic data plus modern post-training methods.

The reason a Q&A session is worth flagging, rather than just another lecture, is what it signals about the audience. The fact that Lambert now has enough reader mail to warrant a second dedicated Q&A, following a first one that covered lectures 1 through 4, suggests the group of people trying to actually implement post-training outside a frontier lab has grown large enough to have real questions of its own.

The book behind the course is not a scrappy self-publish. Per its Simon & Schuster listing it is going through a mainstream publisher, with a Manning edition also available for preorder. The online version at rlhfbook.com stays free, and the whole course, including the Q&A itself on YouTube, sits alongside a codebase and a completion library for anyone who wants to reproduce results.

The honest caveat is that a Q&A is only as useful as the questions it gets, and the course page does not preview which lectures drew the most confusion, how deep the discussion goes on the trickier material, or the size of the reader community driving it. For anyone teaching or hiring in post-training, though, a free canonical reference plus a growing archive of reader questions to point at is a useful piece of scaffolding for a field that is still normalising outside frontier labs.