Every technical field has a story about itself, and that story shapes how the people inside it think. In deep learning the story is roughly that a handful of researchers kept working on neural networks through decades when nobody believed in them, and were vindicated spectacularly. Genius Makers is the reported version of that story, and it is worth reading because the version most people carry around is a compressed myth and this is closer to what happened. The reporting is the book's real credential.
Metz covered this beat as a journalist for years and had access to the people involved, and the result reads like journalism rather than a summary assembled from other summaries. There are scenes and specifics and quotes from people who were there. Anyone who has read a lot of AI writing knows how much of it is the same handful of anecdotes recycled, and this book is a primary source for a fair number of those anecdotes rather than a downstream repetition. What lands hardest is how genuinely unfashionable this work was for a very long time.
It is difficult now, with the entire technology industry reorganised around this technology, to recall that neural network research was a career risk. Papers were rejected. Funding was scarce. Sensible people believed the approach was a dead end and they had reasonable grounds for believing it.
The book conveys the texture of that period well, and it is a useful corrective to hindsight, which makes everything look inevitable. It also complicates the easy lesson about persistence. The researchers were right, but plenty of people persist with approaches that turn out to be wrong, and the book is better than most at showing how contingent the outcome was. The story of how this expertise got acquired by large companies is the section with the most present day relevance.
The auction dynamics, the extraordinary sums, the way entire research groups moved into corporate labs, all of this explains a great deal about why the field looks the way it does now. If you have wondered why so much frontier research happens inside a small number of companies rather than in universities, the answer is largely in this book, and understanding it changes how you read industry news. It is genuinely readable. Metz writes for a general audience without condescension, the pacing is good, and the characters are drawn as people with motivations and conflicts rather than as symbols.
It does not require any technical background whatsoever. Now the limitations, which are significant if you are looking for education rather than context. There is almost no technical content. You will not learn what a neural network does, why depth helps, or what backpropagation is.
That is a legitimate choice for a book about people, but it means this is not a substitute for any technical resource and someone reading it alone will finish with a confident narrative and no understanding of the underlying subject. The publication date matters more than it does for most books. This came out in 2021, before the current era of large language models reached the public, and the field has changed almost beyond recognition since. The book ends before the story most readers are actually curious about begins.
It remains excellent on the period it covers and it will not tell you anything about what happened after. The geographic and institutional focus is narrow. This is largely the story of a small number of labs in North America and the United Kingdom, with the researchers who moved between them. Significant work happened elsewhere, and the Chinese research ecosystem in particular gets far less attention than its contribution warrants.
The book is honest about what it is covering but the effect is a partial picture presented as the picture. The great man framing is the deepest criticism. Narrative history needs protagonists, and building the story around a handful of famous researchers makes for a better book than an accurate account of how research actually proceeds, which is incremental, collective, and built on the work of many people who will never appear in a popular book. The framing is a genre requirement rather than a failure of reporting, but it does distort.
Four point zero for genuinely good reporting on a story worth understanding, told well and accessible to anyone. It will improve your reading of everything else in this space. Marked down for having no technical content, for stopping before the era most people care about, and for a lens that is narrower than the field it describes.