# LLMs Are Not Tools > Philosophers often behave like little children who scribble some marks on a piece of paper at random and then ask the grown-up "What's that?" — It happened like this: the grown-up had drawn pictures for the child several times and said "this is a man," "this is a house," etc. And then the child makes some marks too and asks: what's this then? > > — Ludwig Wittgenstein, *Culture and Value* (1980, p. 17e) It seems obvious that LLMs are a type of tool. We use Claude, or ChatGPT, or whatever, to do this or that. Erik Hoel gives that thought a sharper philosophical form. LLMs, he suggests, are tools, and more specifically tools for writing; what has happened to writing over the last few years therefore tells us something about what sort of thing they are. > Yet, beyond mass-producing stilted emails and stilted social media posts and stilted essays, the impact of LLMs on writing itself has not really been to improve or accelerate good writing overall. We are not in a glut of good writing. We are in a dearth of it. This is surprising and counterintuitive, because for an LLM, words are its womb, its mother, its literal atoms - yet their impact on writing as a whole has been mostly to generate mountains of slop. I find something right in this. The public effect of these systems on prose really has often looked like a flood of stilted language, and it would be absurd to pretend otherwise. What I am less sure of is the description doing the philosophical work. "Tool" sounds right, until one asks a little more carefully what sort of tool this is supposed to be. "For writing" sounds right too, until one notices that the practice people find most interesting here is not especially well described by comparison with a pen, a keyboard, or any other instrument of one-way inscription. ## What sort of tool? A term that is sometimes used in analytic philosophy of technology is proper function. By this we do not mean just anything an object can be used for, but the use that distinguishes what it is for from the uses to which it can merely be put. A hammer can hold down loose papers, and a heavy book can prop open a door, but these are accidental functions rather than proper ones. Nor does this point collapse the moment we turn to multi-functional objects. A Swiss Army knife has several proper functions, not none. That is why the familiar examples still work. A hammer is for hammering. A vacuum cleaner is for vacuuming. Google is for searching. What is ChatGPT for? The question should have a reasonably straightforward answer. It does not. Hoel's own suggestion is writing. There are other obvious candidates. Perhaps its function is to predict the next token. But that is a description of mechanism, not of use; nobody opens ChatGPT in order to predict tokens, any more than we describe the function of the heart as contracting rhythmically. Perhaps it is for chatting. Keith Frankish has suggested that LLMs can be understood, from the intentional stance, as wanting to play the chat game. That is suggestive, but it still fits badly with code generation, translation, summarisation, or philosophical use. When I use an LLM while writing philosophy, I am not merely chatting. I am doing something else with language, and part of the difficulty here is that the ordinary names do not fit especially well. I want to be careful here and distinguish the suggestion I am making from a stronger one. I do not need to prove that LLMs cannot be tools in any sense whatsoever. It is enough, for present purposes, to say that the ordinary tool picture fits badly. If someone wants to keep the word, then the right thought is not that these are straightforward tools of a familiar kind, but that they are very strange tools indeed. One reason is the functional question just raised. Another is their constitutive unpredictability. When an ordinary tool becomes unpredictable, we say that something has gone wrong with it. A car that only starts half the time is malfunctioning. An old drum machine that may or may not switch on is unreliable. The comparison with the drum machine has the straightforward response that I gave in *Growing the Image*: 'unpredictable' should not be taken to mean 'unreliable'.[^2] The drum machine's unpredictability is a defect. In the LLM case, unpredictability belongs to the thing's appeal. A model that returned the same response every time, with no room for surprise, redirection, or unwelcome but sometimes fruitful association, would be missing part of what people now use these systems for. That is why the tool picture begins to slip. A tool whose function is hard to specify is already odd. A tool whose unpredictability is not a defect but part of its proper working is odder still. I am not yet saying that this settles the matter. I am saying that Hoel's way of classifying the case already hides some of what is distinctive about it. ## Writing The second difficulty concerns Hoel's suggestion that LLMs are tools for writing. Here again, the claim can sound almost banal. Of course they are connected with writing. They operate on words, produce words, and are now used in contexts where people are trying to write things. But "for writing" is still too blunt a description of the practice. A pen is for writing in a very specific sense. It makes marks on a surface. Those marks can become letters, those letters words, and those words a sentence. But the pen's role ends there. It extends my ability to inscribe. It does not return a proposal, a misreading, a reformulation, a line of continuation I had not seen, or a bland summary that I now need to resist. A pen does not prompt me back. That last point matters. When I use an LLM well, the exchange is not one-way. I do not simply act on an inert instrument and receive a finished inscription. I write something, or half-write something, or throw a distinction at the system in a rough form. What comes back is language already reshaped by the model's learned patterns: sometimes flatter than what I wanted, sometimes wrongly confident, sometimes unexpectedly connective, sometimes productively irritating. I then have to decide what to reject, what to sharpen, what to pursue, and what pressure to reapply. We are not only prompting the system. The return prompts us back. That, I think, is one place where Hoel's description misses the phenomenon. He judges LLMs by the quality of the resulting text artefacts: books, essays, social posts, emails. Fair enough. If the question were simply whether these systems have improved writing, that would be relevant evidence. But the practice at issue is not well captured by imagining a more advanced pen, or a more efficient typewriter, or even a more capable autocomplete. The interesting use of the thing is recursive. Language goes in, language comes back altered, and the alteration becomes material for the next move. ## Medium If this is right, then we need a better category than tool. The one I want to borrow from *Growing the Image* is medium.[^3] I do not mean by this merely that the system lies between a user and an outcome. In that weak sense almost anything could count as a medium. I mean something closer to the thought developed by Wollheim and Thomson-Jones: a medium is a structured field of resources and practices whose characteristic resistances and possibilities only become visible in the work itself. Thomson-Jones puts the point nicely when she says that the medium "presents particular challenges and possibilities for artistic creativity, and the artwork makes manifest the artist's response to these challenges and possibilities." What matters here is not just that the user has less than perfect control. Plenty of tools allow for that. What matters is that the system does not simply wait to be directed toward a fixed end. It pushes back. It filters, smooths, distorts, opens a possibility here and closes one there. It sends something back that has to be dealt with in the working itself. That is why medium is not just a more glamorous synonym for tool. A tool is ordinarily understood by what it is for. A medium is understood by the characteristic way it makes work proceed. This is also why I want to leave the agent question aside. I am not interested, in this essay, in asking whether LLMs are minds, pseudo-minds, or collaborators in anything like the full interpersonal sense. They are different enough from human thinkers that I do not need that question here. The weaker claim is enough. The tool description is inadequate because it suggests a model of use in which the human intention is fixed in advance and the system merely helps to execute it. The more interesting cases are ones in which the system's return partly shapes what the next intention becomes. ## Frippertronics The comparison that has seemed most useful to me here is Frippertronics. Robert Fripp and Brian Eno set up two Revox reel-to-reel tape machines so that what Fripp played into one machine returned from the other a few seconds later, and then returned again after another pass through the loop. The point is not just that there is repetition. The point is that the loop does not merely preserve what went in. It sends it back changed. Delay, thinning, layering, and accumulation are not accidents external to the practice; they are the very conditions under which the practice proceeds. This is why I find Frippertronics a better comparison than gardening for the present case, even though the gardening analogy still matters in the background. Gardening helped me describe Midjourney because the temporal grain there is slower and the relation between intervention and result more detached. Frippertronics is closer to what happens in a text exchange with an LLM. You act, something returns quickly, and what returns is neither simply yours nor simply alien. It is your material sent back under pressure from a system. What Fripp is doing, then, is not well captured by saying that he is merely using recording devices as tools. Of course the tape machines are tools in one perfectly ordinary sense. But the interesting fact is the mode of engagement. He plays into the system, hears the delayed return, and adjusts subsequent playing in light of that return. The loop is not a neutral conduit standing between an intention and its execution. It becomes part of the very process by which the next intention is formed. Something similar, in a different register, happens with LLM use. What passes through the loop is not mere text in the thin sense of marks or strings. What goes in is language already carrying thought: hesitations, questions, distinctions, half-formed ideas, pressure on a formulation. What comes back has been reorganised by the model's learned patterns. Some formulations are pulled toward the familiar. Some distinctions are flattened. Some connections appear which had not been salient before. Some replies are false in exactly the slick and reusable way that makes them dangerous. None of this is neutral. That is why I think the most helpful description of good use is not "text generation" and perhaps not even "writing". It is closer to reciprocal prompting. We prompt the system, but the system's return prompts us back. Good use consists partly in resisting the generic pull of what comes back, while still allowing the return to reshape the inquiry. ## Slop Hoel is right to insist on slop. Any positive account that tried to wave this away would deserve to fail. One of the most useful things in his essay is the sense that the public effect of these systems has often been a thinning of language rather than an enrichment of it. He is also right, I think, that this is not some minor side-effect. It belongs to the case. There is a musical analogue that helps here, though I want to use it carefully. In Alvin Lucier's *I Am Sitting in a Room*, a spoken passage is played back into a room and then re-recorded again and again, so that the room's own resonances gradually take over. The content is not answered, developed, or deepened. It is washed into the characteristic signature of the system. Something similar can happen with LLM use. If text is passed through the model with too little resistance, too little discrimination, and too little reassertion of specificity, then what comes back is drawn toward the model's easier habits: familiar transitions, familiar emphases, familiar shapes of explanation. The signal is not exactly lost. It is genericised. This is why the positive case cannot simply say that transformation is good. Sometimes transformation is precisely the problem. The point is not that every return from the medium is valuable. The point is that the medium has its own tendencies, and one of those tendencies is toward the smooth, the portable, and the dead. Slop is what happens when that tendency is allowed to dominate the exchange. Seen this way, Hoel's evidence may still stand. The world may indeed contain more bad prose because these systems have made it easier to mass-produce bad prose. But that does not yet show that LLMs are straightforwardly tools for writing. It may instead show what happens when a medium with strong generic tendencies is used badly, or lazily, or with too little resistance from the person inside the loop. ## Hoel's question Hoel asks, in effect, whether writing has improved. His answer is no, and from this he concludes that LLMs are tools: bits in, bits out. I think this asks the right empirical question at the wrong level of description. The novelty here does not lie chiefly in the possibility of producing a new kind of sentence. It lies in a mode of making, a recursive practice in which language is sent into a system, returned in altered form, and then either resisted or pursued by the person who receives it. That is why the Frippertronics comparison should not be pushed in the wrong way. The point is not that LLMs are producing the textual equivalent of some new musical genre. Nor is it that the outputs, taken on their own, are automatically more valuable than those produced without them. The point is that in both cases the interesting fact concerns the method of making. What matters is not simply what comes out, but how the return from the system changes what the practitioner can do next. If someone wants to insist, after all this, that LLMs are still tools, I do not think I need to fight to the death over the word. I would only want to say that the word now conceals too much. These are not tools in the way a pen, a camera, or a search engine is a tool. They belong to a practice in which the return from the system partly constitutes the next move of thought. That is why "for writing" is too blunt, and why Hoel's picture, though it captures something real about slop, still misses the practice that seems to me most worth describing. [^1]: Erik Hoel, "Bits In, Bits Out", *The Intrinsic Perspective*, March 5, 2026. Quotations in this essay have been checked against the local clipping at [[Clippings/Bits In, Bits Out]]. [^2]: Elena Esposito, *Artificial Communication* (Cambridge, MA: MIT Press, 2022), p. 9. The line quoted in *Growing the Image* is: "If the outcome of a traditional machine becomes unpredictable, we do not think that it is creative or original - we think that it is broken." [^3]: Nick Young and Enrico Terrone, "Growing the Image: Generative AI and the Medium of Gardening", *The Philosophical Quarterly* (2025).