AI can give one person the output of an institute, but without sources, that body of work cannot become shared memory
AI can give one person the output of an institute, but without sources, that body of work cannot become shared memory
Cryonicist Michael Darwin noticed NBM House, a nearly invisible publisher that produced more than a hundred essays and short books in roughly 140 days, averaging about nine thousand words a day. He suggests that a human editor may have worked with language models. That hypothesis remains unproven, but it points to a separate problem: readers have almost no way to trace the origins of many ideas and claims.
On July 10, Darwin published an essay about NBM House, which publishes work on philosophy, demography, civilizational collapse, consciousness, and artificial intelligence. He read only part of the collection, but found a recurring method of analysis. The texts first reconstruct how an idea or institution works, then examine its assumptions, limitations, and consequences. The publisher’s public Medium profile has only 43 followers. Its output and range of subjects seem disproportionate to its visibility.
Darwin writes that an individual author usually leaves personal themes and habits in a text, while a large team produces different voices. He attributes NBM House’s consistent method and pace to close collaboration between a person and language models. For now, this is an interpretation of the signs, not an established fact about authorship. In such a collaboration, the person chooses the questions, sets the requirements, and rejects weak lines of reasoning, while the model rapidly assembles context, suggests connections, and helps revise the text.
Darwin is interested in more than the origins of these essays. He has long written about biostasis, the preservation of a person at an extremely low temperature in the hope that restoration will become possible in the future. Such a project depends on organizations that must survive changes in personnel, technology, and governments. These organizations need a body of knowledge that the next researcher can verify, expand, and correct.
This is where NBM House encounters a weakness. Darwin notes the scarcity of citations and the limited visibility of the origins of its ideas. If a text explains why complex societies collapse or which institutions survive a crisis, readers need a traceable path to the data, books, and disputed arguments behind those claims. Without that path, they cannot distinguish a sound synthesis from a convincing error, and the next author must start the investigation again.
Language models remove the former constraint on the speed of reading and writing. Verification still has to be built separately. The medical AI agent ATHENA-R1 demonstrates one such approach by providing a visible trail of sources for its recommendations. For knowledge intended to remain useful for decades or centuries, the length of the collection matters less than the chain connecting a claim, its source, its verification, and any subsequent correction. Without that chain, even a highly productive author remains an interesting interlocutor but cannot provide a reliable basis for decisions on which someone’s future life may depend.