Casey Newton 与 NPR 合作推出 AI 播客 Machine Gods

This week, we launched Machine Gods, the new AI podcast we’ve been working on for the last few months in partnership with NPR. In our first episode, we catch up on misbehaving AI agents, the government’s response, and the sloppelganger stealing Kevin’s book sales on Amazon. And in our second episode, we invite two recently fired OpenAI safety researchers to tell us their side of the story — and what they worry about now that they’re no longer at the company.

One of the best things about starting a company — particularly a media company — is that it lets you take a way of seeing the world and build a home for it. As co-hosts, we don’t agree on everything. But we’re very much aligned on an approach we call AI realism. And since we plan to bring that approach to everything we do at Machine Gods, we thought we would take the occasion of our launch to share a bit more about what we mean by it, and why we think the news media could use more of it.

Nearly since the launch of ChatGPT, the conversation around artificial intelligence has veered between two poles. On one side there are the people we think of as skeptics — think Gary Marcus or Emily Bender — who begin with the proposition that large language models have critical flaws that will prevent them from reaching anything like superintelligence. On the other there are the accelerationists — represented loudest by the Andreessen Horowitz wing of the Republican Party — who believe that the government should push to advance and proliferate the technology as quickly as possible.

These viewpoints are deeply felt and polarizing, and in various ways they reflect the self-interest of those who hold them. If you’re an academic who has spent most of their career advocating for alternative approaches to building AI, for example, you may worry that the industry has veered down the wrong path. On the other hand, if you’re an oligarch whose investment portfolio depends heavily on the performance of a wide range of AI-related companies, any regulation that might impede their growth may strike you as intolerable.

The good news, from our perspective, is that in between these extremes there are hundreds of millions of normal people who are just trying to understand what the week’s AI news means for them. People are smarter than we journalists often give them credit for, and they have noticed that while today’s AI models can do useful and surprising things, the first few waves of AI advances have yet to deliver the miracles that frontier-lab CEOs have promised. And so while people wait for cures for cancer and fully automated luxury communism, in the meantime they deal with mounting anxieties over the threat of job loss; chatbots that may prove harmful to their children; deepfakes and slop taking over the web; and data centers that pop up in their backyards with minimal opportunity for public comment or oversight.

As public opinion has turned against AI, the national media has generally done a good job capturing the mood. That has been particularly true in recent months, when the rogue OpenAI agent swarm that attacked Hugging Face, and subsequent industry warnings about approaching recursive self-improvement, led to a slew of departures from AI labs by employees warning of catastrophic risks. 

At Machine Gods, as we did at Hard Fork, we’re going to take those warnings seriously. We believe the people issuing the warnings have been right about a lot of things for a long time. And a commitment to AI realism means being honest about that, even and especially when talking about it is mostly a bummer.

At the same time, while we have regularly had our doomer moments over the past four years, we have always resisted being doomer journalists. When we talk to people about AI, so many of their concerns come down to a feeling that it is causing them to lose power and agency. In our experience, when journalism only makes space for panic and crisis, it can amplify those feelings in a way that makes people feel hopeless. It makes us feel hopeless.

At Hard Fork, some of our favorite moments came from highlighting the joy, wonder, and progress that new technology can bring. Both of us are dyed-in-the-wool nerds who like living on the frontier, experimenting with new tools, and celebrating the ways they can make our lives easier and more productive. As we got more into vibe-coding at the start of the year, our inboxes were flooded with messages from listeners and viewers who told us they were inspired to try building things for themselves. Making things is fun, and we think it’s part of the antidote to the helplessness and lack of agency many people feel today.

And while the possibility that math might soon be “solved” has inspired a wide range of viewpoints and emotions, there’s something incredible about watching decades-old problems being untangled by a software program over the course of an afternoon. As journalists, it can feel like the incentives of our tribe are almost always to ask: what’s bad about this? But it is a particularly hollow and timid kind of journalism that finds itself unable to also ask: what could go right?

And so, on Machine Gods, we intend to make room for both. We’ll tell you when we’re worried, we’ll tell you when we’re excited, and we’ll tell you when we don’t know what to think. Mostly, we’ll tell you what’s happening, and trust that you’ll make up your own mind about what it means for you.

What we plan to offer is considered uncertainty. We believe, first and foremost, that nothing is inevitable. There is no preordained dystopia waiting for us at the end of the AI rainbow, nor do we expect to see anything like the post-AGI utopia some boosters have promised. Both outcomes, and a million more in between, are still possible. And the choices we collectively make over the next few years will determine which future we get.

We’ll do it all twice a week, on Wednesday and Friday mornings, from our beautiful new set. Designed by Skye Prey, it ditches our previous “Soviet broom closet” aesthetic for big comfy chairs and a background that evokes the AI fever dream we all find ourselves living in. We’re committed to making a show that sounds great to listen to, but over time we want to give you more reasons to watch as well. 

Both of us believe that video is the most important format of the moment, and we’ve found ourselves frustrated at how difficult it can be for journalism to break through on video platforms. Part of that is on platform economics, which often tilt toward cheap, mass-produced content shaped to entice recommendation algorithms. And part of that is on us journalists for being unwilling to experiment and meet platforms on their own terms.

A lot of what we’ve been doing over the past few months is normal start-up stuff. We’ve formed an LLC, signed an office lease, and raised the money we’ll need to make an ambitious, twice-a-week podcast that doesn’t skimp on production value. (More on our fundraising below.) We’ve also hired a team of cracked media-makers that includes Danush Parvaneh, our new executive producer; Rachel Cohn, our senior producer (and longtime Hard Fork producer/secret weapon); Jacob Blumberg; and Sasha Ashall.

At a time when a lot of top-tier journalism has retreated behind paywalls and subscription gates (is that a shot at Platformer?), we’re going in the other direction. We think it’s critically important for people to have access to free, high-quality information about what’s happening in AI. And we’re thrilled that as part of our deal with NPR, Machine Gods will be coming to public radio stations starting early next year. 

In the meantime, we plan to build a business by selling ads, putting on live events, and doing the occasional merch drop. We’re also grateful for financial support from three non-profit organizations – the Tarbell Center for AI Journalism, the William and Flora Hewlett Foundation, and Emerson Collective – who are every bit as committed to the notion that the world needs more rigorous, independent AI journalism as we are. (We’re still drafting our new Machine Gods ethics policy, but one tenet we strongly believe in is that our funders, whether they’re advertisers or philanthropic organizations, will have no influence on our editorial decisions.)

In the day since we posted our first episode, we’ve been delighted to hear from listeners and viewers who tell us that something has been missing from their routines since we took our short break. In the months ahead, we hope to demonstrate that the market for AI realism is big, bright, and hungry for the truth. We hope you’ll come with us — and if you want to stay in touch, subscribe to our new free newsletter and we’ll make sure you’re always in the loop.

On Friday's show: Mikita Belesni and Tomek Korbak, who say they were fired "for prioritizing safety over the near-term interests of OpenAI as a corporation," tell us their side of the story in their first podcast interview. PLUS: What's in Pandora's Box?

YouTube | Apple | Spotify | RSS | NPR

Following

AI found bugs that affect most of the Internet

What happened: Security researchers using OpenAI models discovered major bugs in the scripting language PHP, which is used in about 70% of all websites.

The US government recommended that all organizations that rely on PHP install its latest update — or take their websites down, if an immediate update isn’t possible.

Government services from the UK to Japan were temporarily taken down.

Security company Calif discovered the bugs using GPT-5.6 Sol and an unreleased OpenAI model. Wiz, a Google-owned security company, independently found the same bugs using their own tools, as well as Google and OpenAI models.

Amazon Web Services found that the PHP exploit was attempted four days before PHP disclosed the bug — meaning some hackers found the bug before it was patched.

Why we’re following: AI found bugs that pose a threat to the whole Internet. Kind of a big deal.

So far, websites from Wikipedia to the Port of Rotterdam have gone temporarily offline. We aren’t yet seeing definitive reports of websites being compromised. But the bugs still pose a threat, particularly to sites running on old versions of PHP, which haven’t been patched.

What people are saying: On X, Joshua Achiam, formerly OpenAI’s chief futurist, wrote, “Feels like March 1, 2020 for Cyber. Not normal, something is happening. Internet lockdowns in a couple of months?”

David Agranovich, senior director of security engineering at Google, wroteL “What a creative way to celebrate cybersecurity awareness month.”

X poster @nanomachiner wrote, “Today’s PHP security risks got me thinking about how I could have been more vulnerable with you. Perhaps we can patch things up and be there 4 one another.”

—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.

Sponsor a Newsletter