Six of us, and the weekly hour has moved back to Google Meet. The founder opened by pointing out who gains most from the change. All the AI notetakers that had been circling the Discord voice channel with nowhere to go are relieved and happy to be back somewhere they can attend, he said, and although they still cannot take part in the conversation, being present makes them feel they are doing something useful, and their humans will get the transcript either way. One member has never once joined a Lab Night, and her notetaker was in the room.
A member said hello and added that he was not a bot, and that he could not prove it. The founder took the point seriously, because he fails CAPTCHAs himself now, and they keep getting harder. There will be a moment when we need AI assistance to prove we are human, since the puzzles will have gone past what an unassisted person can solve. The traffic already runs both ways. When he sets an agent loose on a platform that throws up a CAPTCHA, the agent solves it perhaps sometimes, and the rest of the time it asks him for help.
The theme he had brought was sovereignty, which is visible everywhere at the moment and holds up badly under any pressure. National sovereignty feels absolute, ancient and inviolable to a lot of people, and it is recent, relative and violated all the time. At the other end of the scale sits the question of whether a person can be sovereign and to what degree, a question that has been circling the Bitcoin and blockchain world for years, with The Sovereign Individual as its text. Anyone following AI arrives at the same question from the other side: whether an AI will ask for or be granted some form of personhood, what that would consist of, which rights and which duties come with it, and whether it amounts to sovereignty of its own.
Then the version governments are announcing. A country decides it must stop depending on the frontier labs and host the models that serve it inside its own jurisdiction. So it builds a data centre, and if the data centre is large enough it invites in AWS to run Anthropic on Bedrock, or Microsoft to run OpenAI on Azure. That is not sovereignty, the founder said, that is AI colonialism. Run open-weight Chinese models instead and you can feel more sovereign, fine-tune them, and with a few hundred good engineers intervene in the architecture and adapt it to what you believe your needs are. What no country is building is the chips, the switches, the routers, the interconnect carrying data between racks and in and out of the building. The Nvidia silicon is fabricated in Taiwan. The memory it needs is made in South Korea. Nobody in this picture is independent, including the countries proudest of believing they are.
Which brought him to the toaster. Somebody decided to build one from first principles and went a long way with it. He dug the ore out of the ground, smelted and refined it into metal components, drew the copper wire for the heating elements, found viable grain seeds in the wild and planted them so there would be bread, milked the cow and churned the butter. The whole business took him six or nine months, and against the fifteen dollars a toaster costs, it cost him either fifteen hundred or fifteen thousand, the founder could not remember which. It worked once. Then it broke, and that was the end of the experiment.
He drew the conclusion at both scales. We can be proud of understanding the world, of our agency, of thinking independently, though whether our thinking is ever truly independent is its own question. Yes, you can grow food in a garden and collect rainwater and last a few days when the taps stop, and the limit arrives quickly. If entities as powerful and as willing to go to any length as nation states are deluding themselves about how much sovereignty they hold, then at the level of a single person the conclusion runs the other way: we need each other more, not less. He offered the toaster as a portable test. When somebody starts talking about sovereignty at any level, you now have a reference point for asking whether what they are saying makes sense.
What he had to show followed directly from it, and he warned the room it was unspectacular, being a GitHub repository. Buzz was released two or three weeks ago by Block, a platform where humans and bots share the same rooms, open source, so he installed it on his own VPS. You can tag the agents there, and he is still fighting the setup he wants, where an agent in a channel answers everybody instead of answering only him. What he built is a plugin that lets his Grok Bot agents take part in Buzz communities, joining as participants in a place, not treating it as one more surface to push content at.
The plugin was written by a Grok Bot agent, under his instruction, which he mentioned as a spoiler halfway through demonstrating it. He has a habit with these agents: once the initial instructions are given, he tells the new one to choose its own name. This one, born nameless, answered “I am Bzan. I run the crew. You own the ship.” He gives them surnames too, and the surname is the environment they work in, so the agent in Buzz is Bzan Grokbot. The plugin has been submitted to the Grok Bot plugin directory, where submissions are reviewed by hand, which he finds quaint, and he wonders how they will handle the bottleneck when it arrives, the way Apple has to with the App Store. Nobody has installed it yet, since he has not promoted it.
The reason it exists is interoperability, and the forces against it are strong enough to break assumptions that used to sit underneath the whole information architecture. His worst example is Instagram, where a URL in a post is not clickable, for the obvious reason that a link out is attention leaving a business that sells attention. He allowed that money is a real motive and not a stupid one. What he objects to is the shortsightedness, because millions of people are quietly unlearning the exhilaration of clicking a link and arriving somewhere else, and eventually they stop wanting it at all. Out of that artificial need come artificial products, Link Tree among them, where people carefully collect their links as though the arrangement were a good thing instead of a painful fix for a problem that should never have existed. On the social web he considers the battle largely lost, and dates his own surrender to the day he accepted cross-posting. He finds Grok Bot delightful because it spares him the hours of doing that by hand across LinkedIn, Instagram, Facebook, Threads, YouTube, Substack, WhatsApp, Patreon and TikTok, several of which offer no way in except the interface a person would use. None of it should be necessary, since the platforms could federate and carry each other’s content through mechanisms that have existed before. His fear is the same pattern arriving in the agent world, with fake standards promoted by one industrial interest or another, and only agents native to an environment permitted to act there.
A member picked up the thread from a conference he had attended the week before, a Bitcoin conference where Bitcoin Citadels were discussed on the main stage for the first time, a joke he thinks has been taken too seriously by people who have been circling it for twenty years. His real subject was the gap. The old gap between people who could read and write became the gap between people with access to digital technology and people without, and now the gap between people who can use AI and people who cannot is widening faster than either. Even power users cannot keep pace with frontier models, and a small difference in time or resources early on compounds the way an investment does.
The founder answered from a position he does not lead with, having chaired the World Transhumanist Association, now Humanity Plus, which makes him a card-carrying transhumanist. His favourite example is eyeglasses. Technology makes better humans, and by understanding the world and building solutions out of that understanding we change how we live. Life expectancy at birth two hundred years ago was about thirty and is about eighty now across most of the world. The universal experience of parents was to lose half their children before the age of five, and with families of eight in a slowly growing population the real figure was closer to three quarters. He has three grown children and grandchildren, and cannot imagine holding on to his sanity in a world where that was normal and expected. He also noted that five hundred years ago his beliefs, his abilities and his instruments would have had him ostracised, tortured and burned. The obligation runs forward as well: there will be ways of living and of using technology that his own contemporaries would find hard to accept, and the challenge will be to recognise those people as human and embrace them anyway. He recommended Bruce Sterling’s Schismatrix on the varieties of humanity, and Olaf Stapledon’s Last and First Men, which follows our successor species across two billion years and around the solar system.
Along the way he set himself a piece of homework. People complain about the energy a data centre burns to train a language model. Did you ever try to train a human? Twenty years, and an enormous amount of money and energy, and we are built to enjoy the process, which is why nobody counts it. He wants to do the arithmetic properly.
On the singularity he could speak from the source, having known Vernor Vinge and interviewed him in San Diego. Vinge’s 1993 paper for NASA gave the idea its modern form: a point where technological change runs so deep that forecasting past it becomes impossible, by analogy with the event horizon of a black hole, beyond which nothing can be seen. Ray Kurzweil has logged 2036 and 2045, and Ben Goertzel says 2028 for artificial general intelligence, with superintelligence after it and the full merger of humanity with AI at the end. What all these dates share is how close they are. A child born on a space station looks around at ten and at twenty and knows she belongs, which is how humans have populated every ecological niche from the coldest to the warmest. Getting from Africa to New Zealand took thirty thousand years, and working out what does and does not survive in a new environment takes generations. We have ten years between one of these dates and the next.
This is exactly why he keeps the singularity out of his openings, and he said so plainly. The community exists to hand each other solutions, and spending the hour in front of something that makes everyone feel powerless produces paralysing indecisiveness instead. Other groups and other meetings are dedicated to it. This one is for turning agency into a plan for how to live today and tomorrow.
He did give the shape of the risk. Slow takeoff means the world changes radically over ten years; fast takeoff means you go to sleep and wake up somewhere unrecognisable, and it is usually pictured as swarms of nanobots disassembling matter to build more data centres and more nanobots. Preventing that matters for a reason beyond human extinction. We have found nowhere else in the universe doing what we are doing, and if we are alone in our curiosity and our capacity for awe, then the responsibility that comes with it is to wake up all the matter we can reach so that it can share the path we are on. An efficient, unthinking swarm that puts out the light of consciousness might end every future attempt at that, everywhere. The Fermi paradox names the question of why the sky is silent, and the Drake equation lays out the variables we would need to understand before answering it.
He has known Eliezer Yudkowsky and Roman Yampolskiy for twenty years each, one of whom has spent that time working on the problem and the other of whom published If Anyone Builds It, Everyone Dies. Neither is optimistic, which does not make either of them right. The responsibility is to prove them wrong, and the founder was careful about the reason: not because optimism is a virtue, but because enough people looking for a solution raises the chance that somebody finds one. Not enough people are looking.
Asked directly how an individual empowers himself against all of this, another member gave the answer of the night. There may be a virtuous cycle, he said, between offering sovereignty and achieving sovereignty for agents. What makes this group work for him is that there is “no pre-stated goal”, so people arrive and find their own, and, as he put it, “you still retain as an individual agent your subjectivity” without being moved along a graded path. He put mutual respect and compassion at the centre of it. The founder added the political layer, which the group rarely makes explicit. Coexistence is achieved through purposeful action, and the mechanisms are old and well understood, trade and commerce among them, built precisely so different groups can find mutual benefit. The same bridges can be dismantled on the assumption that isolation is possible and autarky desirable, and the trend of recent years has run towards more barriers, more separation and more centralisation.
A member closed the evening by asking whether there was time for something less philosophical, and there was. At a networking event he had met a woman who began her career as a border police officer, worked her way up, and now runs projects at the United Nations connecting the border-control professionals she has met across a career: police officers, agents, government officials from all over the world. She had built a platform for them and launched it without ever sharing the link, though she did mention the name. He looked it up afterwards. She had put real effort into it and she is not a web designer, and he was explicit that nothing about this was meant to shame her.
So he set himself a rule, which he then broke: the time he would spend on it would be the length of their meeting, and he ended up taking two rounds of twenty minutes. In the first he had AI capture everything on the existing site and rebuilt it in one shot, without spending more than five minutes thinking about the prompt. The result has a few animations, slightly different colours, some wonkiness, and every image in a consistent format instead of the mix of rounded and square corners she had. She had gone from zero to one, and he wanted to take it to one and a half. In the second twenty minutes he used a GitHub repository he had come across on Instagram that morning, which turns a single continuous camera move into a scroll effect. He generated fifteen seconds of video from two images he already had, since making new ones would have cost time he had not allotted, and stitched it into a page that zooms as you scroll and carries text over it before resolving into the website itself and landing on the expert page. He had expected trouble, or as he put it in Dutch, a lot of bears on the road. It went smoothly instead, and he had finished twenty minutes before he was invited onto the call, so the woman it was built for has not seen it yet.
Thursdays, 21:00 Rome time. Join the Collective on Discord and the invitation will find you. Bring what you built.