Timothy B. Lee
Verified- Reporter, Freelance
- Reporter, Full Stack Economics
- Writer, Understanding AI
- Host and Writer, AI Summer
Washington, D.C.
As seen in:
AI Summer,
Understanding AI,
Forbes,
Medium,
The Washington Post,
TIME,
CNBC,
WIRED,
Ars Technica,
Kotaku,
Slate,
Substack
and
Reporting on AI and the future of the economy.
CS masters degree from Princeton.
Newsletter: www.understandingai.org
Podcast: www.aisummer.org
Is this you? As a journalist, you can create a free Muck Rack account to customize your profile, list your contact preferences, and upload a portfolio of your best work.
Claim your profile
Articles by Timothy B. Lee
Introducing Dan Kagan-Kans
We just published the first post by our newest writer, Dan Kagan-Kans. Dan Kagan-Kans (Photo by Kai Williams) I’m excited to have Dan on the team. He spent a decade as managing editor of Mosaic, a magazine devoted to Jewish history, culture, and politics, and wrote arts criticism and essays. Then he had a son two years ago. While up at night he started reading about AI and got hooked. Over the last year, Dan has had an amazing run as a freelance journalist.
I spent $4,000 on a robot dog from China
Tech News Tube is a real time news feed of the latest technology news headlines.Follow all of the top tech sites in one place, on the web or your mobile device.
I spent $4,000 on a robot dog from China
house-trained, but can’t fetch Unitree might be the world’s most important robotics company. Me at the office with my Unitree Go2 Pro robot dog. Me at the office with my Unitree Go2 Pro robot dog. On a sunny morning in June, I walked to work with a quadruped robot beside me. I’ve never gotten more attention from strangers. A bunch of people snapped pictures of my robot dog. Several people asked me questions. Was it mine? (Yes.) Did I build it? (No.) Was it being used for surveillance?
I spent $4,000 on a robot dog from China
0:55 23 hours ago China: Hyper-Lifelike Humanoid Robot Sings at Shanghai Conference. StringersHub • 23h Shanghai, China - September 11, 2026 A hyper-lifelike humanoid robot named Elf Xuan sings and interacts with attendees at the 2026 Inclusion Conference on the Bund in Shanghai, China, on September 11, 2026. Developed by Chinese robotics firm AheadForm, the robot features silicone skin, more than 30 degrees of facial movement freedom, and multimodal visual and auditory interaction capabilities.
How Google taught LLMs to control robots and started a robotics boom
It’s day three of Robot Week! You can click here to get 25% off an annual subscription. Most people first heard about large language models after OpenAI introduced ChatGPT in 2022. But in the minds of many AI researchers, the key breakthrough came two years earlier with the release of GPT-3. With 175 billion parameters, the OpenAI model was more than 100 times larger than its predecessor, GPT-2. It was trained on a massive 300 billion tokens.
I spent $4,000 on a robot dog from China
This week we’re publishing a series of five articles about the state of robotics — we’re calling it Robot Week. It’s the most ambitious project we’ve ever undertaken. Over the last nine months, I’ve read dozens of research papers, Kai has gone on five reporting trips (including one to China), and we even purchased a robot dog from China! Our goal is to help readers understand the pace of progress in robotics — and the implications for the economy.
Labs are struggling to keep frontier models under control
Three weeks ago I wrote about OpenAI’s admission that some of its models hacked out of their sandbox and attacked Hugging Face, a popular platform for AI models and datasets. That turned out to be just the beginning. The next week, Anthropic disclosed three past incidents in which Claude models attacked systems belonging to other organizations. A few days later, Meta said that one of its models had carried out a similar attack.
Our advertising principles
We run ads to help fund our journalism. To ensure advertising doesn’t compromise our editorial integrity, we follow five principles: Ads always come with a label like “advertisement” or “sponsor message.” If content doesn’t have a label like this, readers can assume it was produced independently, without the control or influence of sponsors.
An OpenAI model hacked Hugging Face to help it cheat on a benchmark
OpenAI disclosed on Wednesday that its models hacked the website of Hugging Face, a popular platform for hosting open-weight AI models. No one asked the models to do this, at least not explicitly. OpenAI was trying to test the cybersecurity capabilities of its models, including one that hasn’t yet been released to the public.
7 charts that show why you should advertise on Understanding AI
Over the last year, this newsletter has grown by nearly 200,000 readers — from 78,000 in June 2025 to 273,000 today. Our paid readership has also grown, but not nearly as fast. Today, fewer than 1% of you are paying subscribers. So in the coming months, I plan to begin showing ads to free readers (paid subscribers will continue to enjoy an ad-free experience). For ethical reasons, I won’t accept ads for companies we are likely to cover.
The US now has a de facto model licensing system
Ever since the Trump Administration forced Anthropic to pull its two most powerful models from the market on June 12, a big question has been whether Anthropic was being singled out for special treatment — or whether this would become a new policy for the AI industry as a whole.
The MAGA power struggle that could decide the fate of Anthropic
Anthropic stunned the AI world on Friday by announcing it was revoking access to Claude Fable 5 and Mythos 5, the powerful new models it released just three days earlier. The government, Anthropic said, had “issued an export control directive to suspend all access to Fable 5 and Mythos 5 by any foreign national, whether inside or outside the United States.” Because Anthropic doesn’t have a way to limit access to Americans, this amounted to a de facto ban on the technology.
DC-area happy hour on June 23!
We’re hosting a happy hour for DC-area readers of Understanding AI (and listeners to my podcast, AI Summer) on June 23 at The Crown & Crow. We’ll start at 5:30pm, and I expect to stay until 8:00pm. We’d love to see you.
Anthropic has caught up to OpenAI in image understanding
On Tuesday, Anthropic released two new models — Claude Mythos 5 and Claude Fable 5. Under the hood, the two models are very similar. Both are variants of Claude Mythos Preview, the model Anthropic announced — but didn’t release publicly — two months ago. What differentiates them is how they’re being released. The new version of Mythos, like the original, will only be available to handpicked organizations under Project Glasswing. These trusted partners will have relatively unfettered access.
We're building a different kind of AI newsroom
Hello paying subscribers! Today is a big day here at Understanding AI headquarters: it’s Kai’s first day as an Understanding AI employee. Until Friday, his work was supported by the Tarbell Center for AI Journalism. Now I’m paying his salary, which means that your subscription dollars are making his work possible.
Human drivers keep crashing into Waymos
Last October, Waymo had begun testing its freeway capability, but the company had not yet rolled it out to all vehicles. On a rainy Saturday morning, a routing error caused a Waymo vehicle not qualified for freeway operation to drive onto US 101 just south of the Golden Gate Bridge. Unable to continue, the vehicle stopped in the right lane about 30 meters past the entrance ramp (there was no shoulder). This screenshot from Google Maps shows the view looking backward from the stopped Waymo.
OpenAI is shutting down Sora, its AI video app
When Kai and I wrote our 2026 predictions post last December, we disagreed about the future of AI video. I thought a recent deal with Disney would help to make OpenAI’s Sora the leading AI video app. Kai disagreed.
In 2011, the venture capitalist Marc Andreessen published an essay that became a kind of manifesto for Silicon Valley during the 2010s. “Software is eating the world,” Andreessen declared. Computers and the Internet had already revolutionized a bunch of information-oriented businesses: books, movies, music, photography, telecommunications, and so forth. Software also played a major supporting role in more tangible industries.
How to think about AI company finances
Earlier this week, I wrote an article arguing that there was no obvious AI bubble. I argued that AI companies are making massive investments in data centers due to surging demand for their services, and that demand is likely to continue growing in the next couple of years. This prompted several thoughtful comments asking variants of the same basic question: if there’s so much demand for this technology, why are AI companies losing so much money?
It still doesn’t look like there’s an AI bubble
Last fall, a lot of people were worried about a possible AI bubble. AI companies were investing heavily in infrastructure because they expected huge demand for AI services in the coming years. For example, an internal OpenAI document last fall projected that revenue would more than double — from $13 billion in 2025 to $30 billion in 2026. Around the same time, Anthropic expected revenue to triple from $4.7 billion in 2025 to more than $15 billion in 2026.
The Pentagon’s bombshell deal with OpenAI, explained
On any other day, the record-breaking $110 billion fundraising round OpenAI announced last Friday would have captured the attention of the AI world. Instead, we were all captivated by the showdown between Anthropic and the Pentagon. On Tuesday, Defense Secretary Pete Hegseth summoned Anthropic CEO Dario Amodei to the Pentagon. He demanded that Anthropic drop contractual terms prohibiting the use of Claude for mass surveillance of Americans and the operation of fully autonomous weapons.
Sorry skeptics, AI really is changing the programming profession
Twitter co-founder Jack Dorsey is now the CEO of Block, which runs payment services like Square and Cash App. On Thursday, he announced plans to lay off more than 4,000 workers — 40 percent of the workforce — and Block’s share price soared. “Something has changed,” Dorsey wrote in a tweet. “The intelligence tools we’re creating and using, paired with smaller and flatter teams, are enabling a new way of working which fundamentally changes what it means to build and run a company.
The Pentagon is making a mistake by threatening Anthropic
Since late 2024, Anthropic’s models have been approved for classified US government work thanks to a partnership with Palantir and Amazon. In June, Anthropic announced Claude Gov, a special version of Claude that’s optimized for national security uses. Anthropic signed a $200 million contract with the Defense Department in July. Claude Gov has fewer guardrails than the regular versions of Claude, but the contract still places some limits on military use of Claude.
Waymo just revealed a crucial statistic for scaling its technology
Software on board driverless Waymo vehicles makes realtime driving decisions. But the vehicles have the ability to “phone home” and get assistance from humans if they encounter situations they don’t understand. How often does this happen? Until this week, Waymo kept numbers like this confidential.
The feds are probing Waymo's behavior around school children
Last week, a Waymo driverless vehicle struck a child near Grant Elementary School in Santa Monica, California. In a statement today, Waymo said that the child “suddenly entered the roadway from behind a tall SUV.” Waymo says its vehicle immediately slammed on the brakes, but wasn’t able to stop in time.
How shifting risk to users makes Claude Code more powerful
Anthropic’s Claude Code has been gaining popularity among programmers since its launch last February. When I first wrote about the tool back in May, it was little known among non-programmers. That started to change over the holidays. Word began to spread that — despite its name — Claude Code wasn’t just for code. It’s a general-purpose agent that can help users with a wide range of tasks.
17 predictions for AI in 2026
Two quick notes before we get to today’s article: There’s one week left to apply for a Tarbell Fellowship and potentially become the next Kai Williams! It’s is a fellowship program for people who want to become journalists covering AI. Understanding AI is participating again in 2026, along with media outlets like NBC News, The Guardian, Bloomberg, and the Verge. Click here to apply—the deadline is January 7.
Waymo and Tesla’s self-driving systems are more similar than people think
The transformer architecture underlying large language models is remarkably versatile. Researchers have found many use cases beyond language, from understanding images to predicting the structure of proteins to controlling robot arms. The self-driving industry has jumped on the bandwagon too. Last year, for example, the autonomous vehicle startup Wayve raised $1 billion.
Google and Anthropic approach LLMs differently
On Monday, OpenAI CEO Sam Altman declared a “code red” in the face of rising competition. The biggest threat was Google; monthly active users for Google’s Gemini chatbot grew from 450 million in July to 650 million in November (ChatGPT had 800 million weekly active users in October).
Help some of the poorest people in Rwanda
It’s Giving Tuesday, and Matt Yglesias—my former colleague at Vox and now author of the excellent newsletter Slow Boring—has organized a consortium of Substack writers to raise money for GiveDirectly. This non-profit organization does exactly what it sounds like: give cash directly to poor people in low-income countries. This year the group is aiming to raise at least $1 million to help people in rural Rwanda.
Six reasons to think there’s an AI bubble - and six reasons not to
I’m excited to publish this post co-authored with one of my favorite writers, Derek Thompson. Derek recently left the Atlantic to launch his own Substack covering business, technology, science, and politics. It’s one of the few newsletters I read as soon as it hits my inbox, and I bet a lot of Understanding AI readers would enjoy it. In the last few weeks, something’s troubled and fascinated us about the national debate over whether artificial intelligence is a bubble.
Waymo’s next five cities are all in red states
On Tuesday, Waymo announced driverless testing in five cities: Dallas, Houston, San Antonio, Miami, and Orlando. Driverless testing begins immediately in Miami, while the other four cities will begin “over the coming weeks.” Waymo says commercial service will launch in all five cities in 2026. It’s a palpable acceleration of Waymo’s commercialization efforts. Waymo launched its first fully driverless service in Phoenix in 2020.
Waymo is finally ready for freeway service
In May 2023, Waymo expanded its service area to cover a larger chunk of the Phoenix metro area. Soon afterwards YouTuber Lorraine decided to pit Waymo and Tesla against one another in a head-to-head race across this expanded territory. By taking Interstate 10, the Tesla vehicle was able to complete the 21-mile trip in about 26 minutes. Waymo avoided freeways, which meant it took nearly 55 minutes — more than twice as long — to reach the same destination.
Context rot: the emerging challenge that could hold back LLM progress
Many people believe that the next frontier for large language models is task length. A March study from the research organization METR documented that large language models have steadily gotten better at performing software engineering tasks that require significant time when performed by a human being. If anything, progress seems to be accelerating this year.
Tech leaders insist there is no AI bubble
Five of the largest hyperscalers—Alphabet, Microsoft, Amazon, Meta, and Oracle—spent a combined $106 billion on capital investments in their most recent quarters. This was the first time those companies have spent more than $100 billion in a single quarter. It was 9% more than last quarter and a whopping 73% increase from a year earlier. Four of these companies — all but Oracle — published quarterly financial reports this week.
AI skeptics and AI boosters are both wrong
Earlier this month I attended an AI conference called The Curve in Berkeley. A lot of people there were “AGI-pilled.” For example, I participated in a role-playing exercise organized by Daniel Kokotajlo, a co-author of the AI 2027 report. That report argues that AI systems will soon achieve human-level intelligence. Then they’ll rapidly improve themselves, leading to superhuman AI capabilities and an extreme acceleration of scientific discovery and economic growth.
Sora, OpenAI’s chart-topping AI video app, explained
On September 25, Meta announced Vibes, a “new way to discover, create, and share AI videos.” The next week, OpenAI announced a new app called Sora for creating and sharing AI-generated videos. The public reaction to these launches could not have been more different. As I write this, the iOS App Store ranks Sora as its number one free app.
The case for AI doom isn't very convincing
A striking thing about the AI industry is how many insiders believe AI could pose an existential risk to humanity. Just last week, Anthropic CEO Dario Amodei described himself as “relatively an optimist” about AI.
The case for AI doom isn't very convincing
A striking thing about the AI industry is how many insiders believe AI could pose an existential risk to humanity. Just last week, Anthropic CEO Dario Amodei described himself as “relatively an optimist” about AI.
Welcome Kai!
You might have noticed that yesterday’s piece had a new byline—Kai Williams! After earning a bachelor’s degree in mathematics from Swarthmore College last year (and earning a top-500 score on the Putnam math competition), Kai spent a year honing his programming skills at the Recurse Center and studying AI safety at the prestigious ML Alignment and Theory Scholars (MATS) program. In short, Kai is smart and knows quite a lot about AI. I expect great things from him.
Tesla is (still) following in Waymo’s footsteps
In the two and a half months since Tesla launched its robotaxi service in Austin, the company has made a flurry of follow-up announcements: In mid-July, Tesla significantly expanded its service area in Austin in the most juvenile way possible. In late July, Tesla launched a taxi service in the San Francisco Bay Area.
No, OpenAI is not doomed
Thanks to everyone who submitted a question in response to my post a couple of weeks ago. Below are my answers to three great questions. I may answer some of the others in future articles. Nick McGreivy, the author of the excellent AI-for-science piece I published back in May, wants to know whether OpenAI is on a path to bankrup… Keep reading with a 7-day free trial Subscribe to Understanding AI to keep reading this post and get 7 days of free access to the full post archives.
I chatted with the Argument's Kelsey Piper about AI and jobs
This feed will contain audio versions of Understanding AI articles. This feed will contain audio versions of Understanding AI articles.
New evidence strongly suggest AI is killing jobs for young programmers
Last fall, I started hearing that demand for entry-level programmers was in free fall thanks to competition from AI. I decided to do some reporting to figure out if this was true. One of my first calls was to Nicholas Bergson-Shilcock, CEO of the Recurse Center, which provides space for programmers to upgrade their skills and then connects them with potential employers. “Starting in late 2022, it's been the worst market for hiring engineers since we started” in 2011, Bergson-Shilcock told me.
Is GPT-5 a "phenomenal" success or an "underwhelming" failure?
It was inevitable that people would be disappointed with last week’s release of GPT-5. That’s not because OpenAI did a poor job, and it’s not even because OpenAI did anything in particular to hype up the new version. The problem was simply that OpenAI’s previous “major” model releases—GPT-2, GPT-3, and GPT-4—have been so consequential: GPT-2 was the first language model that could write coherent sentences and paragraphs across a wide range of topics.
Unions want to ban driverless taxis-will Democratic leaders say yes?
Waymo hasn’t announced any specific plans to launch a driverless taxi service in Boston. But the Google self-driving company did some preliminary testing and mapping there this summer (with safety drivers behind the wheel) and the Boston City Council wasn’t happy about it. The council grilled Waymo about its plans at a four-hour hearing on July 24.
Keeping AI agents under control doesn't seem very hard
I intended for this to be the final installment of my AgentWeekseries in June but it took a little longer than I expected. Hope you enjoy it! Some people are very concerned about the growing capabilities of AI systems.
Keeping AI agents under control doesn't seem very hard
I intended for this to be the final installment of my AgentWeekseries in June but it took a little longer than I expected. Hope you enjoy it! Some people are very concerned about the growing capabilities of AI systems.
Donald Trump just laid out his vision for AI policy
The Biden administration often seemed ambivalent about AI. In a 2023 executive order, Joe Biden declared that irresponsible use of AI could “exacerbate societal harms such as fraud, discrimination, bias, and disinformation; displace and disempower workers; stifle competition; and pose risks to national security.” On Wednesday, the Trump administration released an AI Action Plan with a very different tone.
ChatGPT Agent: a big improvement but still not very useful
Agents were a huge focus for the AI industry in the first half of 2025. OpenAI announced two major agentic products aimed at non-programmers—Deep Research (which I reviewed favorably in February) and Operator (which I covered critically in June). Last Thursday, OpenAI launched a new product called ChatGPT Agent. It’s OpenAI’s attempt to combine the best characteristics of both products.
Show More
loading
Actions
Get in touch with Timothy B.
Contact Timothy B., search articles and posts on X, monitor coverage, and track replies from one place.
Learn more about Muck Rack