Neuron Makers
First edition · 2026 · 9 parts · 43 chapters

Neuron Makers

The human story of how modern AI was built — from a discredited academic backwater to the most consequential technology of the century. Begin with the foreword.

Part I

A New Kind of Machine

The pre-history: how neural networks were invented, ridiculed, abandoned, and quietly kept alive by a handful of believers until the world was forced to pay attention.

  1. P The Man Who Didn't Sit Down Geoffrey Hinton auctions his three-person company from a Lake Tahoe hotel room in December 2012, days after his students' neural network shattered the ImageNet record. → The moment deep learning broke through and Big Tech realized it had to buy in.
  2. 01 Genesis Frank Rosenblatt's Perceptron in 1958, the Navy's thinking machine, and Marvin Minsky's campaign that buried it. → The origin story and the original sin that sidelined neural nets for fifteen years.
  3. 02 Promise Geoff Hinton, backpropagation, and the connectionist underground that refused to let the idea die. → How a few true believers kept a discredited idea alive against the consensus of their field.
  4. 03 Rejection Yann LeCun, LeNet, the second AI winter, and the years researchers had to hide the word "neural." → The wilderness years, and why so many pioneers ended up Canadian or European.
  5. 04 Breakthrough AlexNet, two GPUs in a bedroom, and the 2012 ImageNet result that cut the error rate in half. → The proof that scale and GPUs change everything — the hinge of the modern story.
  6. 05 Testament Google Brain, the YouTube cat detector, and Jeff Dean's industrial-scale compute. → Why the deep-learning era is also a story about a few engineers who knew how to wield data centers.
  7. 06 Ambition Demis Hassabis founds DeepMind in London with a stated goal of solving intelligence. → AGI ambition arrives as a serious corporate strategy, not a science-fiction dream.
Part II

Who Owns Intelligence?

Big Tech goes all-in. The talent war, the corporate labs, and the moment AI stopped being academic.

  1. 07 Rivalry Mark Zuckerberg cold-calls Yann LeCun; Facebook builds FAIR on LeCun's terms. → How AI salaries exploded and open publication became a competitive weapon.
  2. 08 Hype IBM Watson, self-driving promises, and voice assistants — the marketing of intelligence. → When deep learning escaped the labs and became a marketing department.
  3. 09 Anti-hype Nick Bostrom, Elon Musk's warnings, and the safety-versus-capability tension that birthed OpenAI. → The first serious public worry about AI risk, and how it shaped which labs got built.
  4. 10 Explosion AlphaGo beats Lee Sedol in Seoul in 2016; Move 37; hundreds of millions watch in China. → The most public demonstration of deep learning's power — and the start of the US–China race.
  5. 11 Expansion Deep learning swallows speech, translation, and vision; George Dahl and the statistical old guard fall. → How the technique generalized faster than anyone expected.
  6. 12 Dreamland OpenAI's 2015 founding, its odd nonprofit structure, and the rise of Chinese labs. → The competitive landscape settling into the shape it still has.
Part III

Turmoil

The unintended consequences — what deep learning was actually doing in the world.

  1. 13 Deceit Ian Goodfellow invents GANs in a Montreal bar; the road from synthetic faces to deepfakes. → The generative turn, and the first sign the same techniques could fabricate reality.
  2. 14 Hubris China's AI rise, Baidu's poaching of Andrew Ng, and the Westerners who saw it coming. → The geopolitical reckoning that the AlphaGo match set in motion.
  3. 15 Bigotry Jacky Alciné's tweet, Joy Buolamwini's audits, Timnit Gebru, and the birth of fairness research. → An introduction to AI's bias reckoning — a beginning, not the last word.
  4. 16 Weaponization Project Maven and the 2018 Google employee revolt over Pentagon work. → Where the ethics-versus-pragmatism fight inside the labs broke into the open.
  5. 17 Impotence Content moderation at scale and the limits of AI against an adversary that adapts. → The flip side of the hype: what the technology is genuinely bad at.
Part IV

Humans Are Underrated

Closing reflections on the field as it stood before the world changed — where it was going, and what its leaders believed.

  1. 18 Debate Gary Marcus versus the deep-learning establishment over reasoning, abstraction, and symbols. → The field's most consequential unresolved argument.
  2. 19 Automation Robotics, OpenAI's Rubik's-cube hand, and what lab founders really thought about jobs. → The "what does this mean for work?" question, asked through people rather than econometrics.
  3. 20 Religion AGI as a worldview, and how much of the field's leadership believed they were building a new species. → Crucial context for why OpenAI, Anthropic, and DeepMind behave as they do.
  4. 21 X Factor The forward look written before ChatGPT — the predictions, and the door left open. → The baseline against which everything in the parts that follow can be measured.
Part V

The Transformer

The bridge the earlier story never crossed. Between 2017 and 2022 the field abandoned clever architectures for raw scale, and a single Google paper quietly rebuilt the foundation of everything that followed.

  1. 22 Attention Is All You Need Eight Google researchers invent the Transformer in 2017, name it after a Beatles song, and scatter to found the companies that will challenge Google itself. → The one architectural decision the entire modern era rests on.
  2. 23 The Bitter Lesson OpenAI bets that intelligence is mostly scale and turns a research lab into an API business, half-corrected by DeepMind's Chinchilla while Google flinches at its own chatbot. → How the field learned to stop being clever and just add compute.
Part VI

Detonation

On November 30, 2022, a research preview reached the public and the world reorganized around a chatbot. Inside the labs, the fight over who should build this — and how safely — was already years old.

  1. 24 The Research Preview ChatGPT launches as a low-stakes demo and becomes the fastest-growing consumer product in history within two months. → The day AI stopped being a research field and became a global phenomenon.
  2. 25 Code Red ChatGPT triggers a ten-billion-dollar Microsoft bet, a panic at Google, GPT-4, Bing's unhinged "Sydney," and a rushed Bard launch. → The end of cautious AI deployment and the start of the corporate arms race.
  3. 26 The Schism A dispute over safety, control, and the Microsoft deal splits OpenAI; the Amodeis and a dozen colleagues leave to build a lab premised on the fear that the technology could kill everyone. → Why the field's safety conscience became its own company.
  4. 27 Race to the Top Anthropic tries to prove a safety-first lab can also be a frontier lab, shipping Claude and Constitutional AI while taking billions from Google and Amazon. → The bet that safety and capability cannot be separated.
Part VII

The Reckoning

As the machines reached the public, the wider world tried to take their measure — the warnings, the governments, the courts, and the contest between what AI could fake and what it could discover.

  1. 28 The Warning In the spring of 2023, Geoffrey Hinton quits Google to speak freely and the people who built AI sign a one-sentence statement that it could cause human extinction. → The moment the field turned and warned the public about its own work.
  2. 29 The Blip Over five days in November 2023, OpenAI's nonprofit board fires Sam Altman, the company nearly collapses, and he returns more powerful than before. → Who really controls the most important company in AI.
  3. 30 The Rules From Bletchley Park to the EU AI Act to the veto of California's SB 1047, governments try to govern a technology moving faster than they can legislate. → The collision of a research subculture's fears with real politics.
  4. 31 The Counterfeit and the Real Image and video generation explode into deepfakes, the Taylor Swift fakes, and a wave of copyright suits led by the New York Times against OpenAI. → The first great copyright war of the AI age.
  5. 32 The Nobel Turn DeepMind's AlphaFold cracks a fifty-year-old problem in biology, and in 2024 the Nobel committees honor the people who built neural networks and the people who aimed them at science. → The other face of the same technology: not the counterfeit, but the cure.
Part VIII

The New Machines

Between 2024 and 2026 the models stopped merely answering. They learned to reason, to act, and to use computers, and the industrial base required to run them became the most valuable enterprise on earth.

  1. 33 Teaching Machines to Think A new paradigm of test-time compute lets models reason step by step; OpenAI's o1 and o3 and DeepSeek's R1 send the old benchmarks tumbling. → The second scaling law, and the return of reasoning to AI.
  2. 34 The Agentic Turn Models stop answering and start doing — using computers, writing code on their own, and reorganizing software work around agents. → How the chatbot became a worker.
  3. 35 The Trillion-Dollar Engine Nvidia becomes the most valuable company on earth by selling the one thing every lab needs, and a Dutch-Taiwanese-Korean supply chain becomes the binding constraint on intelligence. → How compute became the scarce resource of the age.
  4. 36 The Supercycle The hyperscalers and OpenAI commit hundreds of billions to data centers, to Stargate, and to a circular web of financing, as the binding constraint shifts from chips to gigawatts. → The largest infrastructure build in history, or the largest bubble.
Part IX

The Frontier

By 2026 the contest had gone global and capital-soaked: open weights against closed, China against the United States, lab against lab, all racing toward a finish line no one could agree was real — carried here up to the present day.

  1. 37 The Weights Get Out Meta open-sources Llama against its own safety team's advice, and the open-weight movement becomes the counterweight to a handful of closed frontier labs. → The fork in the road: should frontier AI be a public good or a guarded asset?
  2. 38 Open Versus Closed Mistral carries Europe's flag, the policy fight over whether weights are too dangerous to release reaches Washington, and Chinese open models begin to lead. → Why "open" became the most contested word in AI.
  3. 39 The Quant Builds a Lab A Chinese hedge-fund founder stockpiles Nvidia GPUs and builds, in two years, the lab that will frighten Silicon Valley and Wall Street. → How the most disruptive AI lab came from outside the system.
  4. 40 DeepSeek Day On a single January day in 2025, an efficient Chinese open model erases roughly a trillion dollars of US market value and puts the entire export-control strategy on trial. → The day the West's assumptions about its lead were challenged in public.
  5. 41 The Talent War Mark Zuckerberg spends billions assembling a superintelligence team, Ilya Sutskever and Mira Murati spin out new labs, and the price of a single researcher reaches nine figures. → When the scarce resource stopped being compute and became people.
  6. 42 The Race to the Present As of May 2026, with Anthropic's Claude Opus 4.8 the newest model on the board, five labs and a Chinese challenger race toward AGI with systems that reason, act, and improve — and no one agrees how close they are. → Where the story stands today, and the question now handed to the reader.