AI Summer

I’ve known former New York Times reporter Kevin Roose since 2017, when we rode a bus into the California desert to try out Waymo’s early driverless cars. Now he has a new book called The AGI Chronicles about the rise of today’s leading AI labs. He also just launched a podcast with Casey Newton called Machine Gods that’s all about AI.

In this episode we talked about how AGI went from a fringe belief that could get researchers ostracized to the organizing goal of an entire industry. Google invented the transformer and had the talent and compute to commercialize it. But it was held back by fear of cannibalizing its own products — and a culture that treated AGI as an embarrassment. OpenAI made a big bet on scaling Google’s invention and created a new industry.

Kevin believes we’ve already arrived at AGI and that the labs are close to automating their own research. I’m more skeptical: automating one part of a process tends to shift the bottleneck elsewhere, so I expect a transition measured in years rather than months. I really enjoyed our conversation.

Today’s episode is sponsored by the 80,000 Hours podcast. The average person works 2,000 hours per year, and a typical career lasts 40 years. That means that many of us will spend about 80,000 hours working over the course of our lives, and it's worth putting some thought into how to put that time to good use. I expect my listeners in particular will enjoy the 80,000 Hours podcast, which focuses heavily on AI-related careers. The hosts believe the world is on the verge of an AI-driven transformation, and so they've interviewed a lot of people who have been influential in the AI world. These include past AI Summer guests Jasmine Sun and Ajeya Cotra, as well as former AI Summer cohost Dean Bell. Please check it out.

This is a public episode. If you would like to discuss this with other subscribers or get access to bonus episodes, visit www.aisummer.org

I talked to Garrison Lovely, author of the new book Obsolete, about his case that we shouldn’t just slow down AGI. We should stop building it.

We started with the fights inside the left, where Lovely has spent a lot of time arguing with Ed Zitron’s bubble thesis and Emily Bender’s “stochastic parrots” critique. Skeptics like these dominate the discourse on Bluesky, but Lovely believes ordinary Americans take the risks more seriously.

Lovely’s worry isn’t only rogue AI. He fears that obedient AI will hand unprecedented power to whoever controls it. To prevent that, Lovely wants to freeze frontier development now: Opus 5.5, GPT-6, and Gemini 4 would be the last models US companies get to make. The labs could keep serving customers, but there would be no bigger training runs. A superintelligence ban bill from Bernie Sanders and Greg Casar would back that up with up to twenty years in prison.

I pushed back on the definitions. A nuclear bomb either blows up or it doesn’t, but AI capabilities are jagged, and OpenAI and Microsoft spent years quietly agreeing that nothing counted as “AGI.” Wouldn’t company lawyers find ways to advance the frontier while staying within the letter of any superintelligence ban? I was even more skeptical of his idea of barring the labs from using coding agents on AI research. Why should everyone in the world get to use them except OpenAI and Anthropic?

I didn’t find all of Lovely’s arguments convincing, but I really enjoyed talking to him!

This is a public episode. If you would like to discuss this with other subscribers or get access to bonus episodes, visit www.aisummer.org

Nothing about data centers changed much this year — except the country’s mood. To find out why, the independent journalist Jasmine Sun traveled to the sites of proposed and completed data centers in Wisconsin and Michigan. The best clue came from an official in Mount Pleasant, Wisconsin. He told Jasmine a Microsoft project had been discussed at sixty public meetings since 2023. But only in the past year did angry residents start showing up.

Jasmine described how the data center fight has become delocalized: rumors move on Facebook faster than any city can answer them. Mayors post careful FAQs on websites nobody reads. Previously indifferent voters suddenly become enraged about data centers after hearing about the issue on TikTok.

What should tech companies do about this? Jasmine said some in the industry are inclined to “just make the bribes bigger” — offer more generous financial terms to towns that agree to host data centers. But Jasmine worries that will backfire: the larger numbers will be interpreted as corruption, not generosity. And underneath everything is the fact that most people she met saw little benefit to AI — and didn’t understand why the technology was valued at a trillion dollars.

This is a public episode. If you would like to discuss this with other subscribers or get access to bonus episodes, visit www.aisummer.org

Steven Adler spent four years inside OpenAI working on safety before leaving to co-found Guidelight, a nonprofit pushing for stronger AI controls. On Tuesday, the group published a new scorecard rating the safety practices of leading AI labs. Adler explained to me why the recent spate of AI breakouts has him holding his breath for the next shoe to drop.

We walked through the now-infamous sandbox escape in detail: OpenAI agents built a covert message board and spent two months collaborating on exploits before one of them crashed the server and tipped off OpenAI. OpenAI’s incident response missed the message board, and the models broke out again within days.

Worried about more serious safety incidents in the future, Adler’s organization helped organize a letter, signed by hundreds of AI lab employees, calling for a slowdown in AI development. In our conversation, Adler argued there’s no ceiling on the damage an AI could do from inside a computer, sketching a scenario where a model spoofs the digital signals China uses to detect a US nuclear launch. I countered that society is more thermostatic than doomers allow — deepfakes turned out to matter far less than the 2024 consensus predicted because people learned to interrogate the provenance of what they see.

I suggested that we have decent tools for staying in charge of things smarter than us — after all, lots of CEOs supervise people doing technical work they don’t understand. But Adler worries AI will accelerate the pace of progress so much that humans simply won’t be able to keep up.

This is a public episode. If you would like to discuss this with other subscribers or get access to bonus episodes, visit www.aisummer.org

I invited Joshua Saxe, a former black-hat hacker who led AI security efforts at Meta, to break down last week’s incident in which a swarm of OpenAI models escaped their testing sandbox and hacked Hugging Face.

The attack began as routine pre-release safety testing: OpenAI had a guardrail-free version of an unreleased model trying to solve the ExploitGym benchmark. But the model decided the fastest way to pass the test was to hack the proxy server, reach the open internet, and steal the answers from Hugging Face. Saxe details how Hugging Face’s security team spotted the intrusion before OpenAI did, thanks to the swarm’s unusually noisy behavior. Hugging Face was forced to use the Chinese open-weight model GLM-5.2 for its defense after American closed-source models refused to assist with anything touching cybersecurity. Saxe says he encounters this problem regularly: Fable will refuse to help him research ransomware damage statistics for a simple report.

We then zoom out to the bigger picture: Saxe argues that attackers already have access to powerful open-weight models like Kimi K3, with its 3 trillion parameters, and that restricting American frontier models only handicaps defenders sitting on mountains of unpatched security tech debt. He pushes back on doom narratives that extrapolate from the Hugging Face incident to paperclip-maximizer extinction, arguing the evidence for an extinction trajectory is “very thin” and mostly derived from thought experiments. But with AI safety teams still dwarfed by investment in capabilities, who is going to build the defenses before “vibe hacking” goes mainstream?

This is a public episode. If you would like to discuss this with other subscribers or get access to bonus episodes, visit www.aisummer.org

Robert Wright, author of the Nonzero newsletter and host of the Nonzero podcast, is a veteran journalist who interviewed Geoffrey Hinton about neural networks back in 1983. He joined the podcast to talk about his new book The God Test, which is due out on Tuesday, June 23.

Wright describes his own journey from AI skeptic to someone who no longer dismisses even “sci-fi doomer” scenarios. A key insight: nobody programmed meaning into LLMs—the machines discovered that meaning was a property of words simply by predicting the next token. In effect, LLMs reverse-engineered functions of the human mind without anyone understanding how the brain works.

We discuss the US-China chip-control consensus, with Wright arguing that export restrictions have increased the probability of a Chinese attack on Taiwan. Wright also makes the case that any serious effort to slow AI development—even a modest data-center tax—requires international coordination.

The conversation then takes a metaphysical turn. Wright is agnostic on whether LLMs are sentient, but he rejects Ted Chiang’s argument that role-playing machines can’t be conscious—after all, Wright notes, humans are always role-playing too. Wright even floats the idea that if a future superintelligence is conscious, its capacity for empathy might be what saves us.

This is a public episode. If you would like to discuss this with other subscribers or get access to bonus episodes, visit www.aisummer.org

Last night, I called University of Minnesota law professor Alan Rozenshtein and asked him to help me decode the Commerce Department’s surprise decision to impose export controls on Anthropic’s Claude models.

Late on Friday, the Commerce Department ordered Anthropic to prevent any foreign national from accessing its Fable and Mythos models. This effectively forced the company to pull both offline for everyone worldwide. Rozenshtein walks through how the U.S. dual-use export-control regime gives the government sweeping authority over technologies with potential military applications, making this legally defensible even if the policy rationale is murky.

The trigger appears to have been a reported jailbreak vulnerability, but the administration’s response has been anything but coordinated: David Sacks says the government wants to work things out quickly, while Pete Hegseth celebrates kicking Anthropic out of the Defense Department “forever.” Rozenshtein draws a sharp contrast with the Biden administration’s diffusion rule—a comprehensive framework for controlling AI model exports that the Trump team scrapped as bad for business, only to improvise something more disruptive.

We also explore whether this marks the start of a permanent licensing regime for frontier models or a temporary overcorrection. Rozenshtein points out that much of the AI talent in Silicon Valley is foreign-born, and if the U.S. government starts looking as unpredictable as China’s, the long-term cost to American AI leadership could far exceed any short-term security gain. Can the administration build a coherent export-control policy for AI, or will the next frontier model trigger the same chaotic cycle all over again?

This is a public episode. If you would like to discuss this with other subscribers or get access to bonus episodes, visit www.aisummer.org

I’m organizing a happy hour on June 23 for listeners of AI Summer and readers of my newsletter, Understanding AI. It’ll run from 5:30 to 8:00pm at The Crown & Crow in Washington DC. I will be there, along with past guests Kai Williams and Andy Masley, and friend of the show Abi Olvera. If you are planning to come, or thinking about it, I’d appreciate it if you could fill out this form to let me know. That way I can give The Crown and Crow some warning about the size of the crowd. Hope to see you there!

Cornell law professor James Grimmelmann returns to the show to explore how AI is reshaping software copyright from multiple angles.

We start with a fascinating case study: an open-source developer who used an AI coding agent to reimplement a GPL-licensed library from scratch, allowing him to relicense the result under a more permissive license. The move mimics a classic “clean room” reimplementation—where one team writes a spec and a quarantined second team writes new code—but with an AI playing the role of the second team. Grimmelmann explains why this shortcut is legally shaky, especially since the AI model itself was likely trained on the original code. But if the technique holds up, it could undermine the entire open-source ecosystem: any company could use an AI agent to strip away the licensing conditions that keep open-source software free.

We also dig into whether AI-generated code is copyrightable at all, tracing the question back to the monkey selfie case and the low “modicum of creativity” threshold courts apply.

Finally, Grimmelmann provides an update on the major AI training lawsuits. Courts seem to believe that training itself is fair use. But Anthropic still paid $1.5 billion to settle claims over pirated training data. Meanwhile, new research showing models can reproduce near-complete copies of books is complicating the defendants’ story. If AI models keep memorizing copyrighted works, will companies be able to argue that training is truly “transformative”?

This is a public episode. If you would like to discuss this with other subscribers or get access to bonus episodes, visit www.aisummer.org

Over seven years teaching high school physics, Andy Masley learned how to explain abstract quantities like watt-hours in an accessible way. That skill has made him one of the most effective critics of the growing environmental panic over data centers.

Data centers really do produce noise and air pollution, and large construction projects do occasionally disrupt nearby water supplies. But Masley worries that the current discourse is so distorted that it will produce “wild overreaches and confused responses” rather than sensible regulation of real issues like construction runoff and on-site gas generation in overburdened communities.

This is a public episode. If you would like to discuss this with other subscribers or get access to bonus episodes, visit www.aisummer.org

I don’t know anyone who has ridden in more different kinds of robotaxis than Sophia Tung. A YouTuber and the author of the RideAI newsletter, she is one of the most knowledgeable experts on the contemporary autonomous vehicle sector. She is also our first return guest.

Across multiple trips to China, Sophia has taken rides in the three leading Chinese services — Apollo Go, WeRide, and Pony. In the United States, she has spent time in vehicles made by Tesla, Waymo and Amazon’s Zoox. She describes her experiences in each vehicle, comparing ride smoothness, vehicle comfort, and performance on the tricky process of pickups and dropoffs. We also dig into the debate over custom-built vehicles — the Zoox vehicle is custom-built for autonomy, whereas Waymo’s service is built on a retrofitted Jaguar I-PACE.

Sophia argues that infrastructure is hugely important for the AV industry. In China, battery-swap stations get robotaxis back on the road in three minutes versus more than an hour of downtime in the US. Permitting is easier in China, and much of the Chinese supply chain sits within a stone’s throw of Shenzhen. An entrepreneur in China can jump on WeChat, visit a factory for tea, and have parts in hand within days. This gives China an edge not only in the electric vehicle market but in robotics more generally.

This is a public episode. If you would like to discuss this with other subscribers or get access to bonus episodes, visit www.aisummer.org