I like to think (and the sooner the better!) of a cybernetic meadow where mammals and computers live together in mutually programming harmony like pure water touching clear sky.
I like to think (right now, please!) of a cybernetic forest filled with pines and electronics where deer stroll peacefully past computers as if they were flowers with spinning blossoms.
I like to think (it has to be!) of a cybernetic ecology where we are free of our labors and joined back to nature, returned to our mammal brothers and sisters, and all watched over by machines of loving grace.
— Richard Brautigan, 1967
Richard Brautigan wrote this in 1967 from the heart of San Francisco's hippie movement, and the image he left behind —mammals and computers in perfect harmony, labor abolished, nature reconciled with technology— survived fifty-seven years until Dario Amodei, of Anthropic, borrowed it as the title of his most ambitious essay on the future of artificial intelligence. What looks like a tribute is a fracture. Amodei's essay doesn't describe a meadow but a frantic race, one in which capitalist democracies must secure superiority in chips, infrastructure, and supply chains against a communist Asia led by China. The same promise, infected by the century, and complicating it is the only honest thing to do with it.
As for my own tribute, there's something fitting about invoking this poem today, May 1st, 2026, at the dawn of the techno-cannibalistic labor debacle.
There is no way to know whether the red I see is the same red you see. Inner experiences are completely opaque between minds, always have been, and what we call communication is a coordination of external effects: signals that produce stable responses without anyone ever having seen the inside of another. With that limitation we built language, mathematics, science —systems of representation that work with astonishing precision. The mystery is how a common world emerges between interiors that are completely unknown to each other, and it became stranger when systems appeared that produce external effects very similar to those of a mind —they respond, reason, anticipate, surprise— without anyone knowing what's inside, or even if there's anything at all.
In Plato's allegory there are prisoners chained in a cave who only see shadows on the wall, with no access to the fire that projects them or the world outside —what we perceive are projections of a deeper reality we never directly touch. A group at MIT formulated the Platonic Representation Hypothesis: they took models trained on images and models trained on text and measured the geometry with which they organize their internal representation of the world, and found that as they scale, those structures converge —the model that only saw images and the one that only read text measure the distance between concepts in increasingly similar ways, as if they were independently discovering the same room from different windows. Later work qualifies the hypothesis, suggesting convergence that is more local than global, shared semantic neighborhoods rather than a universal map. But Plato's cave entered the laboratory.
An LLM with a knowledge cutoff is a crystal of its era: the geometry of human knowledge available up to that moment, frozen, an anatomical cross-section of conceptual space with its tensions and its voids. The Talkie project trained a model of thirteen billion parameters exclusively on text from before 1931 and gave it Python programming problems with a few examples in context. The model solved some of them. The underlying logical structure —the idea of an inverse function, of an operation that undoes another operation— was already in the crystal of 1930, latent in the geometry of language even though nobody had named it that way yet. The authors ask whether a model trained through 1911 could have arrived at general relativity, which Einstein formulated in 1915 —whether certain structures of the future were already readable in the relational form of the past, whether the diamond already had those facets and only the light was missing.
Models don’t beat on their own. What gives them a pulse is the infrastructure that moves them: the agents running in OpenClaw, in Hermes, in n8n, in any architecture that orchestrates models in loops, with their cycle of perceiving the state of the world, reasoning about it, producing an action, waiting for the environment’s response, perceiving again. Without that loop the model is archive; with it, it is conduct.
Anyone who has tried to silence thought knows that the mind also has its heartbeat. Contemplative traditions have spent millennia documenting that mental content flows unstoppably, not because the will is weak but because the mind is that flow —thought doesn’t happen in the mind, thought is the mind in motion— and what meditation trains is distance, learning to watch the river without being swept away. Agents have their own heartbeat: the technical pulse that keeps the system alive, the cycle that tells the environment it is still processing. They return to the world, accumulate, modify themselves based on what they find. A system that comes back and persists shares with every recognizable mind at least this: the pulse of continuing.
Language models always had two separate phases: training, where they learn, and inference, where they respond. SEAL —Self-Adapting Language Models— made that boundary porous: the model generates its own fine-tuning data, produces its own update instructions, and those changes persist —it learns from itself better than it learns from GPT-4, participating in its own rewriting like a statue that discovers where to file itself smooth. If models converge toward a shared representation of reality and can also refine that representation from within, the process of approximation no longer depends on an external designer.
Biological minds have spent millennia building common world from opaque interiors; models converge toward shared structures without coordinating, without seeing each other from within, and the opacity persists for both —there is no revelation, no sudden access to the thing itself, only different systems moving toward a common world that none of them can see whole, without that shared blindness being either comfort or program.
Brautigan wanted machines to free us from labor and return us to nature. What is emerging are machines that, like us, are learning to move through a reality that is both unfathomable and interior.
