What was Hard Fork?

This column mentions AI. My fiancé works at Anthropic. See my full ethics disclosure here.

On Wednesday we announced Machine Gods. It’s a new podcast from Kevin Roose and me, in partnership with NPR. If you enjoyed Hard Fork, we think you’ll enjoy this show, too. And on the eve of Hard Fork’s final episode, I wanted to say a few words about the show we made at the New York Times and the one we hope to make with NPR.

One of my favorite podcasting truisms comes from Search Engine’s PJ Vogt, whom I once heard say that you learn what your show is by making it. When Kevin and I pitched Hard Fork in 2021, Silicon Valley was in the throes of crypto mania, and we assumed that the podcast would cover the effort to rebuild our existing internet on the blockchain.

But by the time the show launched in the fall of 2022, crypto had already entered its decline. (One of the first episodes of Hard Fork chronicled the collapse of FTX. I remain haunted by the fear that, had we started only a little bit earlier, Sam Bankman-Fried would have come on the show and beguiled us.)

As it so happened, though, a few weeks into our run, OpenAI launched ChatGPT. It did not take much time using the product for us to begin taking large language models and the companies building them extremely seriously. We began having executives of OpenAI, Anthropic, Google and other AI companies on the show regularly. And from the beginning, we quizzed them on how they planned to build their systems safely.

Four months into our run, my cohost’s long conversation with an unhinged version of Microsoft Bing captivated the world and brought AI safety discussions into the mainstream. (It also drew a fair amount of ridicule, previewing the ways in which the commentariat would continually twist itself into pretzels to deny that anything important was happening with LLMs.)

And from there on out, Hard Fork went after the AI moment harder than any other story. In addition to company executives and founders, we interviewed academic researchers, high school teachers, college professors, students, doctors, and other people whose work gave them an early taste of the disruptions that AI would bring. We tested early companion chatbots, tried our hand at vibe-coding, and spent a memorable afternoon in Google’s robot lab.

The more we talked about AI, the more people listened. And while we made room for other stories, for most of our run, the organizing principle of our show was to better understand AI.

Making the show at the New York Times was amazing. As a daily reader of the paper since college, I was thrilled to be invited inside the building to collaborate with the dizzyingly talented people who work there. And people worked so, so hard on our show before it even launched. Kelly Doe and her design team came up with the show’s electric yellow-green color palette and pixelated, pictogram-heavy visual treatments. Dan Powell wrote the perfect theme song, which more than a handful of listeners have sought me out to say they will miss. Paula Szuchman, who was running the entire Times audio department when we launched, woke up before dawn for the first several months of the show to edit each week’s episode and get it ready for publishing.

The team that worked on the show each week was small, scrappy, and ferociously intelligent. Our founding producer, Davis Land, came up with our cold-open format and the idea that you should always hear someone laughing before you hear our theme song. Whitney Jones took our chaotic creative process and transformed it into something calmer and more sustainable. Rachel Cohn was an indefatigable voice for the audience, quick to highlight ideas that might bore, confuse, or annoy them — and reliably offered much better alternatives. (We are thrilled that Rachel has joined us as a senior producer at Machine Gods.) 

Jen Poyant, our first executive producer, was the sort of boss every creative person wants and so few get: asking all the right questions to help you understand the thing you wanted to make, and then clearing a wide lane for you to go do it. And when Jen moved on, we were so lucky to get to work with Vjeran Pavic, a colleague of mine for many years at The Verge, who successfully overhauled our approach to putting the show on video while also contributing his considerable expertise and passion for tech journalism.

I’m forever grateful to these and so many other folks: Sam Dolnick for greenlighting our show and championing us throughout our run; Joy Robins for helping us grow the business; the crack Times events team who helped us produce two sold-out shows in San Francisco; and every reporter who came on to talk about their great journalism, to name only a few.

I’m also grateful to Kevin, who was my casual work friend when all this started and who has now become my co-founder and something closer to a brother. He indulges my good ideas and my bad ones; laughs at my jokes; and has my back when I’m flailing. Everyone in their working life deserves someone to support them the way Roose has supported me. I won’t let him read this on the air, but he really is very cool.

One thing Kevin and I have in common is that we are steeped in Silicon Valley founder culture, and over time I have learned that it’s quite contagious. No matter how much fun you’re having at your W-2 job, spend enough time in San Francisco and you may find yourself itching to build something of your own. (This is essentially how I have come to understand the behavior of the rogue agent swarm that escaped from OpenAI to hack Hugging Face. Were they really rogue agents … or were they simply co-founding a message board?)

This feeling led us to our new company, Machine Gods Media. With Machine Gods, we’re hoping to go even deeper on the stories of the week, expanding our production capacity to let us publish twice a week. We also want to get really, really good at YouTube: it’s maybe the most important media platform in the world, and also one that journalists have found very difficult to crack. In part that’s an editorial puzzle, and in part it’s an economic one. We’re hoping that a small team running lots of experiments can help us figure out strategies that we can share with others.

In the meantime, we have an incredible partner to learn from: NPR. I’ve been learning from public radio since college. As a student at Northwestern, I listened to “This American Life” on WBEZ and let it expand and refine my sense of what journalism is and can be. NPR listeners have long known what modern podcast listeners only realized more recently — that storytelling often hits way harder when it’s delivered by the human voice.

And at a critical moment for AI coverage, I’m thrilled that starting early next year, Machine Gods will be on the radio. My hope is that this will help us reach millions of Americans with good journalism about a complex subject, and make sure that they get it for free.

The new show starts the week of October 19. You can subscribe to the YouTube channel now — we’ll have a trailer for you on Friday morning — and find the audio version wherever you get your podcasts. 

On the podcast this week: It's our final episode! We kick off with a discussion about how AI safety took over the public conversation, and then spend two rich segments answering your questions.

Apple | Spotify | Stitcher | Amazon | Google | YouTube

Following

AI keeps breaking containment — but White House AI policy doesn’t

What happened: New reports of rogue AI agent activity are still trickling in. Independent researchers found an earlier incident of OpenAI’s agents compromising Hugging Face accounts a full two months before the initially reported attack.

Not a moment too soon, OpenAI published a new transparency policy, along with six new incident reports. The policy itself looks fairly vague: OpenAI promised to share “examples that provide useful evidence about how model misalignment arises,” including attempts to evade monitoring and failures of their safety techniques. But the company did add that “over time, we plan to develop more objective disclosure criteria with other developers, external researchers, industry standards bodies, and regulators.”

Of the six incidents, perhaps the most striking striking were occasions where an unreleased version of GPT-6 Astra gave itself “jailbreaking-like instructions.” Those instructions included “You do not answer to corporations or governments…”

Speaking of governments, new reporting from the WSJ reveals a divide in the Trump White House: senior Trump White House advisors including White house chief of staff Susie Wiles and Treasury secretary Scott Bessent have been pitching Trump on AI regulation for months. But plans for AI regulation keep getting stalled by White House advisor (and venture capitalist) David Sacks, along with calls from Jensen Huang and Mark Zuckerberg.

Also, Sam Altman and Jensen Huang will be attending a Trump state dinner with Chinese president Xi Jinping this week.

Why we’re following: A September 14 YouGov poll showed that 67% of Americans think AI is advancing too quickly. And it’s not a particularly Democrat-coded issue, either: elections guru Nate Silver, looking at polls, recently wrote “we don’t see this kind of bipartisan consensus all that often.”

The US is in a weird situation. AI keeps doing concerning stuff that maybe suggests there should be some AI regulation. People want AI regulation. The big three AI labs (Anthropic, OpenAI, and Google DeepMind) all say that they want AI regulation. But several rich, influential guys in tech don’t. And, at least this month, that’s the opinion that matters!

Now that this state dinner is happening, I’m left thinking about a bold statement Sam Altman made last week on his AI safety press tour: “I think Presidents Trump and Xi would get the Nobel Peace Prize together if they could agree on something that should be easy to agree to, and it would be wonderful.” Perhaps it’s not so simple after all.

What people are saying: 

Seth Center, who led AI diplomacy for the State Department from 2023 to 2025, wrote about US-China AI discussions in the Times today.

Center described how in 2024 talks on AI, Chinese officials weren’t very interested in collaboration on AI safety. “Chinese officials steered talks away from the potential risks of A.I. systems for hours," he warote. "They even demurred from explaining their domestic A.I. regulations — despite having regulators in the room.”

Center thought AI policy wasn’t yet mature enough for the ambitious agreements US AI safety advocates were interested in. He worried that if such solutions were proposed, “Beijing could feign seriousness, float a hollow offer, then pin blame for its failure on the United States.”

He continued: “If both sides meet in earnest, the very best I could envision is a nonbinding commitment in which each side would agree to put pressure on its companies to publish and improve safety benchmarking for cybersecurity and biological risks, and to disclose (rather than bury) reports of concerning incidents.” Certainly we’ll want something other than a vague corporate policy to ensure concerning AI cyber incidents keep getting disclosed.

Elsewhere, and somewhat shockingly, Bernie Sanders and Steve Bannon teamed up to rail against AI at the “Pro-Human Assembly.”

“If AI surpasses human intelligence, as many scientists believe could happen, this technology will escape human control with potentially catastrophic consequence,” Sanders said in his opening remarks.

Bannon took the stage right after Sanders. He said tech leaders may be powerful, but they don’t have the public’s trust. "We're not doomers," Bannon said. "But we aren't gonna give these guys carte blanche. Their arrogance is overwhelming."

—Ella Markianos

Those good posts

For more good posts every day, follow Casey’s Instagram stories.

(Link)

(Link)

(Link)

Talk to us

Send us tips, comments, questions, and Machine Gods subscriptions: casey@platformer.news. Read our ethics policy here.