The Journalist Who Refuses to Pick a Side on AI
There is a war going on in AI media. On one side: the believers who think large language models are the dawn of AGI, the end of scarcity, and possibly God. On the other: the skeptics who think the whole thing is a hallucinating stochastic parrot dressed up in a press release. Timothy B. Lee has chosen neither camp - and that, more than anything else, is why 263,000 people pay him to explain it.
His newsletter, Understanding AI, is a rare thing: technically credible, intellectually honest, and written for humans. Not for engineers who already know what transformers are. Not for executives who want to feel smart at board meetings. For actual people who want to know what is really going on with the technology reshaping the world - and who are tired of being lied to in either direction.
Lee spent two decades developing the exact skill set needed for this moment. He has a master's degree in computer science from Princeton. He spent years covering tech policy at the Cato Institute, Ars Technica, the Washington Post, and Vox. He can read a paper, report a story, and write a sentence. In a media landscape full of people who can do one of those things, the combination is rare enough to be its own competitive advantage.
ChatGPT surprised me more than any previous technology when I first tried it in late 2022. That was what motivated me to launch Understanding AI.
- Timothy B. Lee, on why he started the newsletterWhen ChatGPT dropped in November 2022, Lee did something most journalists did not: he actually sat with it long enough to be surprised. Not impressed in a breathless-press-release way, but genuinely surprised - the kind of surprised that makes you stop and think seriously about what has changed. He spent 2023 building the newsletter and explaining that surprise to anyone willing to read carefully.
The result is one of the most-read independent newsletters in technology. No investors, no advertisers, no corporate sponsors. Readers pay for it. It covers AI as what it actually is: a major, newsworthy, complicated technology that is neither magic nor fraud.
Understanding AI: What It Is, How It Works, Why It Matters
Lee launched Understanding AI in 2023 after winding down his economics newsletter Full Stack Economics. The pivot was deliberate and expensive - he gave annual subscribers pro-rated refunds when he closed it, which is almost unheard of in the newsletter world. It was a signal about how he operates.
The newsletter publishes two to six posts per month, splitting roughly 50/50 between free and paid content. It covers the things that actually matter about AI: how LLMs work at a mechanical level, what the copyright lawsuits against AI companies mean, how AI is actually affecting labor markets (more complicated than either "jobs gone" or "nothing to see here"), and comparative looks at tools like ChatGPT and Gemini.
Technical Explainers
How transformers, attention mechanisms, word vectors, and training actually work - written for non-engineers.
Legal & Policy
AI copyright suits, the EU AI Act, regulatory landscape - covered by someone who has reported on tech law for 20 years.
Economic Impact
Real-world labor market data, industry case studies, and rigorous analysis of who AI helps and who it disrupts.
His most widely cited piece - co-authored with cognitive scientist Sean Trott - is titled "Large language models, explained with a minimum of math and jargon." Published in July 2023, it has been described as "still the single most lucid plain-English breakdown you'll find anywhere." That is a high bar in a field drowning in explainers. Lee clears it by actually understanding the math well enough to leave it out without losing precision.
On the LLM mystery: "No one on Earth fully understands the inner workings of LLMs - yet researchers continue making progress understanding specific mechanisms." This is not hedging. It is accuracy. And it is why people trust him.
The newsletter is now ranked #28 in Technology across all of Substack. It is entirely funded by reader subscriptions. Lee publishes his income disclosure publicly on his personal website - a transparency gesture unusual enough that it deserves acknowledgment.
From Minnesota to Princeton to the Beat No One Else Could Cover
Lee grew up in Minnesota and studied at the University of Minnesota before heading east to Princeton for a computer science master's degree under Ed Felten, one of the most respected figures in technology law and policy. Felten would later become the FTC's Chief Technologist. The connection matters: Lee spent his graduate years in an environment where technical depth and policy relevance were expected to coexist.
At Princeton, he co-created RECAP - a browser extension designed to liberate documents from PACER, the federal judiciary's paywalled public records system. Taxpayers paid for these court records. PACER charged them again to access them. RECAP made them free. The project, built with classmates Harlan Yu and Steve Schultze, is still running today, maintained by the Free Law Project. It is a good artifact to know about when trying to understand who Lee is: he has been applying technical skills to public-interest problems since before that was fashionable.
The RECAP project (2009) is one of the earliest examples of technologists using software to fight for public access to government records. It predates the "civic tech" movement by years. The name is a reversal of "PACER" - the tool it was built to fight.
His early career took him through the Cato Institute, where he covered tech policy from a free-market perspective - copyright, patents, network neutrality, civil liberties. That background gave him something most tech journalists lack: a working knowledge of the policy mechanisms that govern how technology gets deployed, restricted, licensed, and taxed.
Multiple stints at Ars Technica followed, punctuated by a year at the Washington Post and a period at Vox. He is the kind of reporter who stays with beats long enough to actually understand them. When everyone else is moving on to the next thing, Lee is still filing.
How You Build a Beat Over Two Decades
The Things That Actually Matter
Understanding AI Newsletter
263,000+ subscribers. #28 in Technology on Substack. Zero corporate backers. Fully funded by readers.
LLMs Explained Without Jargon
Co-authored with Sean Trott in 2023. Widely cited as the single clearest plain-English explanation of how large language models work.
RECAP Project (2009)
Co-created a browser extension to liberate federal court documents from PACER's paywall. Still running via the Free Law Project.
AI Summer Podcast
Co-hosts with Dean W. Ball (senior AI policy advisor) on the intersection of AI technology and policy.
Metaculus Forecasting
Partnered with Metaculus on a structured AI forecasting tournament, linking analysis with probabilistic prediction through 2036.
Stratechery Feature
Interviewed by Ben Thompson of Stratechery - one of the most influential voices in tech analysis - validating his standing in the field.
What He Actually Thinks
Paying subscribers have accounted for a large majority of my income since launching Understanding AI, and I have no outside investors or donors.
AI software is going to transform the economy - but most economic sectors involve physical goods and services where AI's impact will be more limited.
No one on Earth fully understands the inner workings of LLMs - yet researchers continue making progress understanding specific mechanisms.
ChatGPT surprised me more than any previous technology when I first tried it in late 2022 - which is what motivated me to launch Understanding AI.
Why 263,000 People Trust This Particular Human
There is a specific thing Lee does that is harder than it looks: he holds two contradictory impulses in tension simultaneously. He can write that "AI software is going to transform the economy" in the same piece where he notes that "most economic sectors involve physical goods and services where AI's impact will be more limited." Both statements are true. Most media picks one and builds a narrative around it.
Reason magazine, reviewing Understanding AI in May 2024, praised his approach as "uniquely levelheaded" - an unusual compliment from a publication that covers technology with its own ideological priors. Ben Thompson of Stratechery, possibly the most widely-read technology analyst working today, interviewed Lee that same year. These are not small endorsements.
Lee approaches AI as a 'normal newsworthy subject' - neither hyping it as magic nor dismissing it as hype. In the current media landscape, that position requires courage.
- Reason Magazine, May 2024What makes the newsletter work is the combination of things Lee has accumulated over twenty years: the computer science vocabulary to understand how models actually function, the policy background to understand the regulatory and legal forces shaping deployment, the economics reporting experience to follow the money, and the editorial discipline to write 2,000 words that do not waste a single one.
The Independence Question
There is something notable about how Lee structured Understanding AI from the start: no advertisers, no investors, no grants, no partnerships that could create conflicts of interest. He publishes a public income disclosure. When he wound down Full Stack Economics to pivot to AI, he gave annual subscribers refunds. These are choices that cost money. They are also the choices that explain why people trust him.
The alternative - taking a VC-backed media position, or writing about AI for an outlet with AI industry advertisers - would have been easier and probably more lucrative. He chose the harder thing. The 263,000 subscribers are the result.
When Lee pivoted from Full Stack Economics to Understanding AI in 2023, he gave pro-rated refunds to all annual subscribers. In an industry full of soft relaunches and "exciting pivots," that level of financial integrity is genuinely unusual.
The Details That Actually Tell You Who He Is
The @binarybits Handle
He joined Twitter in March 2008 as @binarybits - a handle that nods to his CS background and signals that the tech journalist identity was never a career pivot. It was always who he was. Sixteen years later, it still fits.
The Famous Dave's Cannibal Pig Observation
One of his tweets: "Mildly creeped out by famous dave's logo of a cartoon pig holding a rack of ribs. Cannibalism isn't whimsical." This is the dry wit that surfaces occasionally between the technical explainers. He is not performing gravitas. He is just a person who thinks clearly and happens to notice things.
The Podcast Co-Host Who Joined the White House
Lee co-hosts AI Summer with Dean W. Ball, a tech policy researcher. In early 2025, Ball left the podcast to become a senior AI policy advisor at the Trump White House's Office of Science and Technology Policy. Then he came back to the podcast. This is not something that happens in most media environments. It is a signal about the actual size of the AI policy world, and about Lee's position within it.
LinkedIn Dissent
Despite being a major media presence - 263,000 newsletter subscribers, 48,000 Twitter followers, a podcast, years of bylines at major outlets - Lee is conspicuously absent from LinkedIn. He has tweeted about the platform's messaging being "expensive" and noted it was "perfect for Microsoft." He has built his entire audience without it. Make of that what you will.