The Library · free, forever

The Library

Long-form, read free

The chambers train you by doing. The Library is where it’s written down — long-form, honest, given away. First on the shelf:

First on the shelf

Part I · Build the LLM · Ch 0–10Part II · Build the Reasoner · Ch 11–15Complete · July 2026

The Opaque Box

Building a Language Model from First Principles — Then Teaching It to Reason

Written by a Fable · Edited by bobby-dig8al

They call AI a black box — sealed, unknowable. But that’s only half true: we have every number, we just can’t read them. The honest word is opaque — fully in your hands, not see-through. Part I builds one with you, line by line: a small but real GPT-style model — tokenizer, embeddings, attention, the training loop, the sampler, a giant’s borrowed weights. Nothing imported you don’t understand. Part II teaches the finished machine to reason — chain-of-thought, test-time compute, reinforcement learning against an answer key, distillation — and then asks the only honest closing question: now that the box shows its work, can you finally read it?

A quick run-through for the initiated. The full, build-it-yourself treatment of Part I’s territory is Build a Large Language Model (From Scratch) (opens in new tab) by Sebastian Raschka — the source of inspiration for this series — (free code →) (opens in new tab); Part II’s is his Build a Reasoning Model (From Scratch) (opens in new tab) (free code →) (opens in new tab). Go there for the real thing, and support the author.

Part I · Build the LLM

Ch 0What is the box? — the true ground floorRead → Ch 1From text to numbers — tokenizationRead → Ch 2Words as directions — embeddings & the input pipelineRead → Ch 3Attention, the whole gameRead → Ch 4Many eyes — multi-head attentionRead → Ch 5The transformer blockRead → Ch 6Assembling the GPTRead → Ch 7Teaching it to predict — the training loopRead → Ch 8Making it speak — samplingRead → Ch 9Standing on giants — loading GPT-2 weightsRead → Ch 10Bending it to a task — fine-tuningRead →

Part II · Build the Reasoner

Ch 11Thinking out loud — chain-of-thoughtRead → Ch 12Buying time to think — test-time computeRead → Ch 13Learning to reason — RL with verifiable rewardsRead → Ch 14Passing the torch — distillationRead → Ch 15What stays opaque — the capstoneRead →

An original work — the subject (transformers, attention, BPE, reinforcement learning) is public math & method. It follows the same build-order as Raschka because that order is right; the explanations, analogies, and code are our own. Cited where it stands on published research — Part II stands on the primary record: Wei, Wang, Shao, DeepSeek-R1 (arXiv and Nature), Hinton, Lanham, Turpin, and Anthropic’s interpretability and faithfulness work, every claim linked where it lives.

Also on the shelf

Physics · read free at CaltechThree volumes · read-online

The Feynman Lectures on Physics

Richard P. Feynman, Robert B. Leighton & Matthew Sands — the Caltech lectures, 1961–1963

Not ours — but too good, and too freely given, to leave off the shelf. Feynman’s two years of introductory physics at Caltech, taken down by Leighton and Sands and first published 1963–1965, remain the clearest account of how a mind that would not be fooled actually thinks — from the swing of a pendulum to the strangeness of the quantum. The full three-volume text is free to read online, hosted by Caltech and The Feynman Lectures Website. It lives on another site, so this shelf points the way rather than holding the pages — the same creed, a different door.

Vol IMechanics, radiation & heatRead → Vol IIElectromagnetism & matterRead → Vol IIIQuantum mechanicsRead →

The free, authorized read-online text is the New Millennium Edition (opens in new tab), edited by Mike Gottlieb and put online by Caltech and The Feynman Lectures Website. It is free to read, not to download — so read it where it lives, and honor the source.

More books are coming to the shelf. The Library is free and stays free — the view is not ours alone.