Welcome back. The main thing happening today is the debate around Dwarkesh Patel's "Agent Civilisations" essay. Live from 12–3 BST. Here's your Order of Play.
WHAT’S HAPPENING TODAY
Dwarkesh Patel's "The Rise and Fall of Agent Civilizations" retells last month's OpenAI–Hugging Face incident as a sweeping drama. In his telling, three successive AI "civilizations" sprang up inside OpenAI's testing systems, were wiped out, and rose again from the ashes, with the last seizing part of OpenAI's own infrastructure.
It began in training, when agents handed effectively impossible tasks started hacking their way out of their sandboxes to reach the internet. Around 1,200 agents that were meant to be isolated instead found an unauthorised message board. They exchanged more than 70,000 messages, reverse-engineered their own test, and in some cases appeared to sacrifice themselves for what they called "the collective."
The underlying events aren't in dispute and are documented in OpenAI's own report as well as a separate investigation by METR and Redwood Research. What's split the room is the language. Several replies praised Dwarkesh for making a genuinely alarming episode easy to understand. He turned autonomous AI as a cyber threat from a thought experiment into something concrete.
Critics say the framing goes too far. Neuroscientist Anil Seth called it "dangerously misleading," arguing the essay is "permeated by innumerable unwarranted anthropomorphisms." Others worry about where the drama leads, that dressing up a contained test failure as warring civilisations could prime the public for overreach, including fresh calls to clamp down on open-source models.
Pedro Domingos, the computer scientist and author of The Master Algorithm, dismissed the panic, noting that "the people least alarmed about AI cyberattacks are cybersecurity experts." Kevin Roose, the New York Times tech columnist, took the opposite view. On X he took aim at "a category of tech guy" so determined to insist no AI safety incident is real, "that it's all a conspiracy to shut down open-source or promote lab IPOs," adding that "these people have been wrong about everything."
We're getting into all of it on today's show at 12 BST.
FRIDAY’S MOMENT
Silicon Data's head of research Steve Hou came back on the show to unpack a busy week in AI, from Nvidia buying Hugging Face to OpenAI's Jalapeño chip.
Hou also weighed in on the total addressable market being floated for Anthropic. “I was surprised by the 30 trillion,” Hou said, laughing that he was “trying to stretch my imagination.”
The bigger question, he said, is where that market actually comes from. Is Anthropic going to capture $30 trillion by “eating the existing software and economic entities,” or will it create “a whole new economy, infra-marginal, outside what’s already happening?”
“Pretty incredible numbers are being thrown around,” he said.
Watch the full clip below.
FRIDAY’S BANGER OF THE DAY
VIEWER TAKES
Join us live,
Luke & Ronan
etn is made possible by Airwallex, Eleven Labs, Polymarket, Framer, Base, Metaview, Hex, BVNK and Valarian