Writers have always imagined someone on the other side of the page. For most of history, that someone was another human being: a friend, a patron, an editor, a student, a future scholar, or simply an unknown reader who might someday understand. Writing was an act of communication, but it was also an act of preservation. A thought placed into language could travel farther than the body that produced it.

Now another kind of reader has entered the world. It is strange, powerful, unreliable, tireless, and increasingly involved in the work of searching, summarizing, comparing, and organizing human knowledge. In online AI culture, this reader is sometimes jokingly called the shoggoth, a reference to the Lovecraftian monster that became a meme for the alien model beneath the friendly chatbot mask. The image became popular in AI circles in late 2022 and early 2023, especially as a way of talking about the difference between the underlying model and the user-facing system shaped by reinforcement learning from human feedback.  

The meme is useful, but the deeper point is not the meme. The deeper point is that writers now inhabit a world in which human beings are no longer the only readers that matter. AI systems already read, compress, retrieve, and synthesize text. They are not wise judges yet, and they are certainly not perfect historians. But they are becoming part of the intellectual environment. Anyone who writes online is increasingly writing into a world where future machine systems may help decide how ideas are found, connected, attributed, and remembered.

This has already led to a small literature about “writing for AI.” Gwern has argued that ordinary human writing may become more important if it can influence or inform future language models, especially when it is cleanly available and readable on the open web.   Scott Alexander has described several versions of the idea, including helping AIs learn what one knows, presenting arguments that future AIs might take seriously, and leaving enough writing behind that an AI could model a person or their views.   Dan Kagan-Kans, in The American Scholar, frames the issue as a possible future in which humans still write, but increasingly do so for AI readers as well as human ones.  

Those are interesting possibilities, but I think there is another reason to write for this new reader. It is less about influence, persuasion, simulation, or immortality, and more about provenance. Future AI systems may become the best available tools for reconstructing the history of ideas. They may be able to ask who said something, when they said it, how clearly they said it, what evidence they offered, what earlier work they drew from, and how their claim relates to later developments. If that becomes possible, then writing online is not merely a way of broadcasting a thought. It is a way of leaving evidence.

This connects to what I have elsewhere called the Final Library. In that essay, I argued that AI will increasingly become both the mechanism of expression and the mechanism of invention. Machines will not merely write up ideas that humans already have; they will search idea-space, test combinations, generate hypotheses, connect distant literatures, and eventually assemble a repository of human and machine-generated knowledge too large and complex for unaided human beings to navigate directly. The Final Library is not simply the internet, or Google Scholar, or the latent space of one language model. It is the imagined end state of an increasingly comprehensive, AI-mediated map of ideas, arguments, sources, hypotheses, and conceptual relationships.  

If something like the Final Library is coming, then the problem for independent thinkers changes. The question is no longer only, “How do I get this idea accepted by today’s institutions?” The question also becomes, “How do I leave this idea in a form that future systems can find, understand, compare, and attribute?”

The game has changed in another way as well. It is not only easier to preserve an idea; it is easier to examine one. A person outside academia can now ask a large language model or deep research system to search for adjacent theories, identify the closest existing literature, generate objections, find missing terminology, and explain how a specialist might criticize the argument. These systems are not authorities, and they are not substitutes for expertise, experiment, or formal peer review. They can miss important sources, misunderstand technical details, and sound more certain than they should. But they provide a new layer of pre-review. They allow many more people to ask whether an idea is original, whether it has already been proposed under another name, whether it is obviously wrong, and whether it deserves to be written down.

This matters enormously. In the past, a person outside an institution had very little access to the ordinary machinery of intellectual feedback. If you had an idea in psychology, neuroscience, physics, economics, anthropology, or philosophy, how would you know whether it was novel? How would you know which field owned the question? How would you know which papers to read, which terms to search, which objections mattered, or which assumptions would immediately bother a specialist? For most people, these questions were almost impossible to answer without mentors, libraries, advisors, seminars, conferences, journal clubs, and professional networks.

Now a first-pass version of that process is available to many more people. It is imperfect, but it is real. LLMs do not replace peer review; they democratize pre-review. They give independent thinkers an intellectual mirror. They can say, “This resembles adaptive calibration theory,” or “This sounds like predictive processing,” or “Someone may have already called this allostasis,” or “The weak point in your argument is that you have not distinguished novelty from deprivation.” They can help a person discover that an idea is not new, which is useful. They can also help a person discover that an idea may be new, or at least underdeveloped, which is even more useful.

This creates a new pathway for ideas. The old pathway was something like: enter an institution, find a mentor, specialize, obtain credentials, receive funding, present at conferences, publish in journals, build a citation network, and eventually become discoverable. That pathway still matters, especially for empirical research. But it is no longer the only path by which a serious idea can enter the record. A newer pathway is becoming possible: think carefully, use AI to test and contextualize the idea, write it clearly, publish it online, archive it, and let future systems discover it if it proves valuable.

This does not make every idea good. It does not mean a blog post is equivalent to a peer-reviewed paper. It does not turn speculation into knowledge. What it does is lower the barrier to responsible intellectual participation. More people can now investigate whether their thoughts have predecessors. More people can locate the relevant conversation. More people can receive criticism before publishing. More people can preserve their ideas in public form. That is a major change in the ecology of thought.

It is also a change in the ecology of recognition. Human institutions have always been imperfect memory systems. They are not only engines of truth; they are also social networks. Ideas travel through advisors, departments, journals, funding agencies, conferences, friendships, rivalries, review articles, and fashionable terminology. These channels are often useful, but they do not guarantee that the best ideas will be seen.

Papers get seen when they are attached to already-active citation clusters, prominent labs, fashionable terminology, review articles, conferences, advisors, grants, and repeated follow-up publications. A paper can be conceptually strong and still fall into a kind of citation shadow if it is not picked up by one of those networks. But that does not mean the paper failed intellectually. It means it failed socially, institutionally, and distributionally.

I have had this experience with my own work. In 2016, I published a paper called “Chronic stress, cortical plasticity and neuroecology.” The paper argued that chronic stress-induced changes in the hippocampus, prefrontal cortex, amygdala, and caudate might constitute a neuroecological program: a shift away from time-intensive, explicit, controlled, top-down processing and toward faster, implicit, automatic, bottom-up responding under adverse conditions.   The point here is not personal complaint. The point is that many ideas live at the edges of existing networks. Some are undercited papers. Some are independent essays. Some are blog posts. Some are PDFs on personal websites. Some are written by people who do not have an institutional home.

A useful metaphor is the fossil. A fossil is not failed life. It is preserved life. It may sit underground for a very long time, unseen by any living mind, but its value is not determined by how quickly it is discovered. Its value depends on whether it remains intact enough to be interpreted later. An idea can be like that. Today it may get no traffic. It may not be cited, discussed, rewarded, assigned, or reviewed. But if it is written clearly, dated, named, and preserved, it becomes evidence. It becomes something future readers can examine.

This is why the Internet Archive matters. The Wayback Machine allows people to save web pages, and the Internet Archive explains that saved pages can be cited, shared, linked to, and preserved even after the original page changes or disappears. Its “Save Page Now” tool creates a permanent URL for a saved page, although it saves a single page rather than an entire site.   That may sound mundane, but it is profound. It gives independent thinkers a simple way to create public, timestamped evidence that a particular text existed at a particular time.

An archived page is not peer review. It is not proof that the idea is true. It is not a guarantee that everyone will agree about priority in every possible dispute. But it is a public timestamp. It gives future humans and future machines something to inspect. It says: this idea was here.

That is enough to change the emotional meaning of writing from the margins. A blog post does not need to become famous immediately to have value. A post can function as an intellectual specimen: labeled, dated, preserved, and available for future classification. The open web exposes the artifact. The Internet Archive preserves it. AI systems help search and compare it. The Final Library, if something like it emerges, may eventually place it in a larger museum of ideas.

This is not search engine optimization. It is not an attempt to game AI systems by stuffing pages with keywords or writing in a flattened machine-friendly style. In fact, writing for future AI should also be good writing for future humans. The goal is clarity, not gimmickry. State the core claim. Explain the problem. Say what is new. Name the predecessors. Link to earlier versions. Use stable terminology. Define unusual terms. Distinguish speculation from evidence. Archive the page. Make the intellectual structure visible enough that a future reader can reconstruct what was being claimed.

This is especially important for people who think across fields. Institutions are usually organized by discipline, but many ideas are not. An independent thinker may notice a pattern between neuroscience and ecology, between machine learning and cognitive psychology, between mythology and evolutionary theory, between economics and thermodynamics. Such ideas are often difficult to publish through normal channels because each field may see only the part that belongs to someone else. AI systems are imperfect, but they may eventually be better at recognizing conceptual overlap across domains than human institutions have been.

There is also a humane point here. It should be possible to contribute an idea without devoting one’s entire life to the institutional promotion of that idea. Traditional academic recognition often rewards people who can build a whole career around one research program: the first paper, the follow-up papers, the review articles, the grants, the students, the conferences, the lab, the citation network. That is a powerful model, and it has produced much of modern science. But not every thinker works that way. Some people generate many ideas across many domains. Some are independent. Some are too interdisciplinary for a department. Some have no realistic access to grants, advisors, conferences, or journals. Some simply want to think and write.

For such people, the new advice is simple: put the idea into the record. Use AI to challenge it. Ask what has already been said. Ask where it is weak. Ask what would falsify it. Ask what literature it belongs to. Revise it. Then publish it with a clear title, a date, your name, a direct thesis statement, and enough context that a future system can understand what you meant. Save it in the Internet Archive. Keep the archived link. Do this calmly and without grandiosity. The purpose is not to claim victory over the future. The purpose is to leave a well-formed trace.

We should not assume that AI will be fair by default. It may inherit the biases of the web, the academy, language, prestige, and citation density. It may overvalue what is already visible. It may miss obscure work, just as human scholars do. But we can make fairness easier by leaving better evidence. A clear, archived, attributable idea is easier to recover than an ambiguous memory, a private notebook, or a buried social media thread.

The democratization of publication came first. Anyone could put words online. The democratization of preliminary intellectual review is happening now. Many more people can use AI systems to test, contextualize, and refine their ideas before publishing them. The democratization of attribution may come next. If future systems become better at mapping conceptual history, they may be able to notice that an idea was anticipated in a neglected article, developed in a blog post, rediscovered in a later paper, and popularized somewhere else.

That would not be a perfect meritocracy. But it would be a better memory.

This is the hopeful possibility. The internet made it possible to publish without permission. The Internet Archive made it possible to preserve without permission. AI may eventually make it possible to be rediscovered without permission. None of this removes the need for rigor, humility, evidence, criticism, or revision. But it does mean that independent thinkers have more reason than ever to write their ideas down.

A fossil buried in the ground is not nothing. It is waiting for the right excavator. A blog post sitting quietly on the internet may be the same kind of object. It may not be read today. It may not be cited tomorrow. But if it is preserved, future systems may one day know where to place it. The museum is still under construction, and the Final Library is not yet here. But the sediment is forming now.