Interconnects AI · 2026-08-10 · notable
Nathan Lambert's RLHF book ships — Manning prints the free post-training text
Nathan Lambert's post-training textbook, Reinforcement Learning from Human Feedback, is now shipping in hardcover from Manning. The full text stays free to read at rlhfbook.com, alongside a codebase and a 12-hour course.
The free RLHF book is now a printed Manning hardcover, with a codebase and a 12-hour course beside it.
What is it?
Reinforcement Learning from Human Feedback: Aligning and Post-training LLMs is Nathan Lambert's textbook on how modern language models are trained after pretraining, and Manning began shipping the hardcover on August 10, 2026. The complete text stays free to read online at rlhfbook.com. Manning is selling it at half price with the code PBLambert through August 19; UK copies arrive in October.
How does it work?
The book is organized around the canonical RLHF recipe and walks through each optimization stage in order: instruction tuning, reward-model training, then rejection sampling, reinforcement learning, on-policy distillation and direct alignment algorithms. Lambert covers policy-gradient methods including PPO, GSPO and CISPO, and devotes space to the systems problems practitioners actually hit — data freshness, training-inference mismatch and throughput in asynchronous setups.
Why does it matter?
Post-training know-how mostly circulates as scattered papers and lab folklore, which makes it hard to enter the field without sitting inside a frontier lab. Lambert gathers it into one ordered reference and keeps it free online, so a researcher can read the whole thing before deciding whether to buy the print copy. Companion resources — a codebase, a completion-comparison library and a 12-hour course with slides and video — turn the book into a teachable curriculum rather than a single text.
Who is it for?
ML researchers and engineers learning post-training
Try it
Read the full book free at https://rlhfbook.com