5 useful things you'll learn in my new post-training textbook (shipping now!)
| Source: Interconnects (Nathan Lambert)
Tags: RLHF, post-training, LLM alignment, Nathan Lambert, reinforcement learning, textbook
Nathan Lambert has published 'Reinforcement Learning from Human Feedback: Aligning and Post-training LLMs' via Manning—covering rejection sampling, outcome reward models, and on-policy distillation with a free online version, 12-hour YouTube course, and 50% discount (code PBLambert) until August 19.
Details
Nathan Lambert, known for the Interconnects newsletter and his research on open model post-training, has released a full textbook on RLHF through Manning Publishing. The book targets practitioners with a CS background who want to understand post-training rigorously—covering topics Lambert says had little accessible documentation when he started: rejection sampling, outcome reward models, character training, and on-policy distillation.\n\nThe book is available three ways: purchased from Manning (50% off until August 19 with code PBLambert), on Amazon, and freely online. A 12-hour course including slides and YouTube video accompanies the book, along with a code repository and suggested exercises.\n\nLambert positions this as the resource he wished existed when he was getting started in post-training. It emphasizes intuitions and historical context over pure derivations, covering the techniques that have remained core to industry post-training pipelines even as nomenclature has evolved. The title acknowledges being slightly outdated—RLHF as a term has partially given way to broader post-training framing—but Lambert added a last-minute section on on-policy distillation to keep the content current.\n\nWith post-training expertise in high demand as labs compete on alignment and instruction-following, this book fills a gap in practitioner-focused documentation for a field that has largely existed in blog posts and papers.