#draft #substack # The Dereliction of Thought* In a recent piece, [[Freddie deBoer]] expresses what I take to be a pretty common sentiment about generative AI in 2025: _%%add hyperlink to piece%%_ > a lot of people who call themselves writers appear to spend most of their days talking themselves into believing that having ChatGPT do their work is still ‘really writing.’ I am one of those people. I am a philosopher: I do a lot of writing, I use LLMs to do it, and I am pretty convinced that what I am doing is really writing. Worse, I am pretty sure that I use LLMs to *think*, and that this is really thinking. This is what I intend to write about here: AI, writing, and thinking. ## Why this exists In December 2022, I was miserable and bored. Although I'd recently been offered a new research position, the bureaucrats at my new institution had decided (I can only assume as a joke) to make it as hard as possible for me to actually begin work. My last position had finished a few months ago, and the money was starting to run out. I was kind of preparing for the new position, but my heart wasn't really in it. Instead, I spent a lot of my days messing around with [[ChatGPT 3]].5, which had been released at the very end of November. Pretty much immediately I was hooked. This was *fucking cool*. I’ve been into tech since my dad bought my brothers and me a Sinclair ZX Spectrum +2 when I was five –I also have very dim memories of playing a Bruce Lee game on my dad's green screen, DOS computer). Over I'm the person who is over eager to help you choose your next phone or laptop, and tuts if they see you writing in Microsoft Word. In 2022, generative AI was the new tech on the block, it fit [[the pattern]]. At the same time, there's a distinctive giddiness to my excitement around AI that I certainly haven't felt for a while. I got a kick out of sending a message and watching text –text typed by no one– spill across the screen.  I was born in the early 80s, and those big-budget Hollywood sci-fi films that we all watched were typically about one of, or a combination of, three different subjects: aliens, time-travel, AI. So far, no Xenomorphs and no Biff Tannen (kind of), but we kind of have that last one now! Kind of! The kid inside me, the one who watched Terminator/Blade Runner/RoboCop/Short Circuit/etc. on repeat, really can't understand why people aren't more excited.  Two clarifications. First, let me qualify that 'Kind of'. To be clear, I don't think that LLMs are conscious, or that they are persons, or person-like or anything like that. My position on this will get clearer over the next few pieces I write, but [[the idea]] that these things have minds, strikes me as, frankly, bananas. I am not saying that my mind is unchangeable, but for the time being I won't be writing about this sort of stuff. **Second, don't mistake my infantile giddiness for boosterism. we should be sceptical of the companies that make them, and worry about what could go wrong.  How worried we should be, and, what, specifically, we should be worried about, are not topics about which I have anything interesting to say. There are plenty of people to read who talk about just this sort of stuff.** ### What I am going to write about I want to write about how I use AIs, specifically LLMs, because so many colleagues, and friends, and people I read on Substack (like [[Freddie deBoer]]) *hate* AI. I am not deny that there is plenty of stuff to hate, there is a truly hideous banality to a lot of LLM generated text. But, I don't What this enthusiasm has led to is me spending a great deal of time thinking about LLMs and the other types of generative AI that have arrived in the last few years. Quite quickly I started to throw philosophy at chatgpt, claude etc.: individual questions, texts by me, texts by other people, and, as of August 2025,  I am happy to say that I _do_ philosophy _with_ LLMs.  When I tell people this, a reasonably common response is worrying about whether these things are smart enough, and whether they— and the fact that these things hallucinate. Just briefly: "smart" is a funny word here, because it's something we attribute to persons or animals, not to token‑prediction machines. But let's say they are very capable of producing text that is indistinguishable at the sentence and paragraph level from real philosophy written by real people.  Again, I’ll say more about this in future posts. The more interesting — at least to me — response I get is dismay.  I suspect the root of this dismay is something like the following. When I say 'I do philosophy with LLMs', what I'm actually saying is 'an LLM is doing [[the philosophy]] for me'.  There is something to this : it is already perfectly possible for savvy students to generate a reasonable essay, one that doesn't feel like it 'was written by AI', and would lightly get a good grade from an unsuspicious teacher, with minimal effort, with a prompt or two. This is no small problem for educators, and If that’s what I were doing you could rightly accuse me of not [[doing philosophy]],  For the exact same reason that it doesn't seem right today, the school student did his homework. in the same way that a school student didn’t do his homework. But that’s not what I’m doing. ## Am I Kidding Myself? - It doesn't seem like I am. - It still feels like work. - It's not as though I'm typing in "do some philosophy" and then leaving the room to come back later. - Until about 2023, much of my working life was spent staring at a screen. - Now I am staring at ChatGPT, Raycast, Gemini, or several chats used concurrently instead of Google Docs. - Am I doing the philosophical work or am I just a cheat? - I don’t care much — I am interested in philosophical ideas regardless of origin. - My feelings of intense relaxation about using LLMs are related to this attitude. - Over the course of a morning, I might have several LLM chats on various topics, often working through my own ideas. - During this process, an LLM may produce a particular expression of an idea or a connection I hadn’t thought of before. - We should not underestimate these systems’ capacity to produce reasonably well-thought-through, well-grounded philosophical ideas. - This is to be expected if you consider how LLMs function. - Of course, LLMs can and do hallucinate. - Nonetheless, a particular idea generated by an LLM would not have emerged without my prompting. - Prompting well is essential for these systems to give you good ideas. - I feel a degree of ownership over all the ideas that have come out of LLMs when I’ve been using them. - Why This Substack - Although this feels like I'm really thinking and writing, part of the reason to start writing this Substack is to explore these ideas. - I want to work out whether I'm right. - I want to work out what it means to think and use [[artificial intelligence]], especially LLMs, to do interesting, thoughtful work. - Another motivation for making this Substack: - Many of the people I read online and like are, if not AI doomers, then deeply dubious — brackets "filled with doob," as we used to say when I was a kid — about [[the value]] of AI to people and to the world generally. - One of them is Freddie de Boer, quoted at the top of this article. - Others include Oliver Berkman and John Elledge, who also seem, yes, not happy with the way things are going. - I am not really trying to argue against these sorts of ideas. - One hope I have is that by showing how I use LLMs, I can show people [[not just]] the possibilities. - Remember, I am not an efficiency or productivity blogger. - I also want to try and convince these people that good humanities-flavoured research, work, and writing can be done with LLMs. - A philosopher friend recently told me that, in the same way he isn’t very interested in AI-generated music or images, he wouldn’t be interested in AI-generated philosophy. - This reflects an interesting split: does philosophy need to be done by people? - My intuitions lean one way, but I want to explore and push against them to test them. - What This Blog Is - Although I will talk about how I use LLMs extensively, the aim of this blog is not to give weekly tips on how best to use ChatGPT. - Someone interested in replicating my techniques might learn something, but I am not a productivity blogger. - I intend to write about what interests me. - The practicalities of how I use these tools interest me. - I am a philosopher interested in the theoretical aspects of LLMs. - I am not a computer scientist, but I have a working knowledge of how LLMs work. - I can tell my [[gradient descent]] from my autoregression. - I am learning more about the technical aspects of these systems. - I am also a long-time tech enthusiast. - My first excitement over technology was for the minidisc player I bought with paper round money around 1997. - For 25 years, I have been the person who thought hard and talked about his phone or laptop. - Coming across AI in 2023 was a continuation of this love of tech. - AI Arrival - AI excited me in particular. - I can’t understand why others are not as wild-eyed. - In the 80s and 90s, science fiction had three big themes: aliens, time travel, and robots. - We now have robots (AI). - We should be as excited about AI as we would be if aliens landed or a time traveller arrived. - What I Am Not - I am not an AI booster. - I do not think AI is unequivocally good for the world. - I do not see people like Sam Altman as prophets sent from heaven. - This is not me saying AI is wonderful. - My focus is my personal interest and fascination. - I will tell you how I do philosophy with AI, but I am not saying everyone should do this. - [[This works]] for me and nothing more. - Although this feels like I'm really thinking and writing, part of the reason to start writing this Substack is to explore these ideas. - I want to work out whether I'm right. - I want to work out what it means to think and use [[artificial intelligence]], especially LLMs, to do interesting, thoughtful work. - Another motivation for making this Substack: - Many of the people I read online and like are, if not AI doomers, then deeply dubious — brackets "filled with doob," as we used to say when I was a kid — about [[the value]] of AI to people and to the world generally. - One of them is Freddie de Boer, quoted at the top of this article. - Others include Oliver Berkman and John Elledge, who also seem, yes, not happy with the way things are going. - I am not really trying to argue against these sorts of ideas. - One hope I have is that by showing how I use LLMs, I can show people [[not just]] the possibilities. - Remember, I am not an efficiency or productivity blogger. - I also want to try and convince these people that good humanities-flavoured research, work, and writing can be done with LLMs. - A philosopher friend recently told me that, in the same way he isn’t very interested in AI-generated music or images, he wouldn’t be interested in AI-generated philosophy. - This reflects an interesting split: does philosophy need to be done by people? - My intuitions lean one way, but I want to explore and push against them to test them. > It feels to me like AI might do something similar for the human brain. The sort of stuff I write – factual but intended to entertain, rather than just inform – requires reading lots of stuff and seeing what grabs me. Sure, I can ask an AI to give me a bunch of facts about [insert topic here] – but it won’t spot the striking coincidence, the funny detail, the historical anecdote I never knew was there but which I suddenly realise is the framing device for my entire story. > > Jonn Elledge ["What cars did to your city, AI will do to your brain"](https://jonn.substack.com/p/what-cars-did-to-your-city-ai-will) **This series tests that fear against concrete workflows.** **Two background commitments guide me. First, names a workflow in which judgement, selection, and responsibility remain mine. Second, names a workflow in which I cede those roles. This series is about the first.** **LLMs are tireless at reorganising and restating; they are weak at selecting what matters; philosophy depends on the latter.** **Worked example. I gave an LLM a rough paragraph from a paper on auditory perception. It collapsed a distinction I needed to keep. I replied: “Preserve the two‑way distinction; do not invent literature; keep premises explicit.” The next draft respected the structure but misweighted the objection. I then asked: “Weight by inferential role. Which premise does the work?” The revision aligned with my aims. The judgement in each step was mine; the model supplied options I filtered.** I am an analytic philosopher. A significant portion of my job is writing articles to submit to philosophy journals.  ** In the past I’ve written about auditory perception and the experience of time; currently most of my work is focussed on AI (insert hyperlink to paper here).**  Something that sits behind these sorts of fears is the idea that if, by using an LLM, the LLM is doing the philosophy rather than the philosopher, then you’re not really doing it any more. And there’s something to this. In August 2025, it is fairly easy for students who want to have an LLM write an essay to get a good grade, as long as the teacher is not suspicious. They can do this pretty easily with one or two prompts, as long as they’re clear about what they want. And, yes, if that were all I did when I was using an LLM, then you’d say I’m not doing philosophy – in the same way that the student didn’t write their homework. But that’s not what I’m doing. The post I’ll put up simultaneously with this introductory one will make a start in explaining how I use LLMs and what this can tell us about thinking and working. However, this Substack should not be mistaken for a 'how to make money with AI' Substack. I’m certainly not going to tell you how to monetise LLM skills or anything like that – that’s not my wheelhouse at all. Although I will draw on technical details of LLMs, I am not a computer scientist. I have a working understanding and can separate out the gradient descent from the autoregression, but I don’t think that’s going to be particularly useful for readers at such a low level. What I am going to try to do is explain how I use them, and also explain why I find them fascinating and what I think this says about how we think and work. ### What this is / is not - A record of methods for thinking and writing with LLMs. - Not a monetisation playbook or 'how to' prompts newsletter. - Not a debate about machine consciousness; the focus is practice. I was born in the early 80s, so a lot of my childhood and teenage years were spent watching the blockbuster science‑fiction movies of the 80s and 90s. It seems to me there were three epics those movies were about: aliens, time travel, and AI. And we have that last one now! Maybe I’d be more excited if aliens landed, but I can’t understand why people aren’t more excited that we’re getting closer to AI like in the movies. Two clarifications. I am not an AI booster; I do not know whether AI will be an unalloyed good, and I do not plan to write policy takes here. Also, when I say we feel close to 'Hollywood AI', I do not mean these systems are conscious or persons, nor likely to be so in the near future. My focus here is on workflows and thinking practices. _(aside: The weird thing in philosophy, especially with technology: if somebody is writing a paper about philosophy and virtual reality, or aesthetics and virtual reality, would you expect them to know the latest models of VR headsets? In parts of aesthetics, at least, this doesn’t seem to be expected by people whatsoever. I’m also looking at how colleagues and friends of mine use LLMs. Again, it seems quite far removed from how other people might use them.)_ _Next up: Interrogating texts with an LLM._ **With one worked example and the limits of the method.** **The Dereliction of Thought*** **In a recent piece %%add hyperlink to piece here%%, Freddie deBoer expresses what I take to be a pretty common sentiment about generative AI in 2025:** a lot of people who call themselves writers appear to spend most of their days talking themselves into believing that having ChatGPT do their work is still ‘really writing.’ \ I am one of those people. I am a philosopher: I do a lot of writing, I use LLMs to do it. I have convinced myself pretty damn thoroughly that what I am doing is really writing. Worse, I am pretty sure that I use them to write  I am pretty damn convinced that I use , I am someone who uses large language models (LLMs) to write. Not only that, I use them to think. Most likely, if you think that writing with LLMs isn't really writing, you're very unlikely to think that thinking with LLMs is really thinking. I started this Substack because I'm interested in questions that surround these sorts of ideas. I'm interested in authorship and creativity. I'm interested in writing. I'm interested in thinking. **This newsletter asks what changes in thinking and authorship when a philosopher collaborates with LLMs.** **This project examines authorship, workflow, and cognition in the presence of LLMs, with methods from analytic philosophy and practice from day-to-day writing.** **Why this exists** I am an analytic philosopher. A significant portion of my job is producing text for philosophy journals. **My job is reading, drafting, revising, and defending arguments for journals and talks.** In the past I’ve written about auditory perception and the experience of time; currently most of my work is focussed on AI (insert hyperlink to paper here). Since the arrival of GPT-3.5 in November 2022, I’ve been pretty much obsessed with generative AI. I've always been into tech – the sort of person who will ask you what phone model that is – and AI is the new technology on the block, so it certainly fits the pattern. At the same time, there's a giddiness to my excitement and interest in AI that I don't remember having about any other sort of technology. _[Aside: outlet for AI thoughts without burdening friends or full papers.]_ What this enthusiasm has led to is me spending a great deal of time thinking about LLMs and the other types of generative AI that have arrived in the last few years. Quite quickly I started to throw philosophy at ChatGPT, Claude, etc.: individual questions, texts by me, texts by other people, and, as of August 2025, I am happy to say that I _do_ philosophy _with_ LLMs. When I tell people this, a reasonably common response is worrying about whether these things are smart enough and about the fact that they hallucinate. Just briefly: "smart" is a funny word here, because it's something we attribute to persons or animals, not to token‑prediction machines. But let's say they are very capable of producing text that is indistinguishable at the sentence and paragraph level from real philosophy written by real people. Again, I’ll say more about this in future posts. The more interesting – at least to me – response I get is dismay. I suspect the root of this dismay is something like the following. When I say 'I do philosophy with LLMs', what I'm actually saying is 'an LLM is doing the philosophy for me'. It feels to me like AI might do something similar for the human brain. The sort of stuff I write – factual but intended to entertain, rather than just inform – requires reading lots of stuff and seeing what grabs me. Sure, I can ask an AI to give me a bunch of facts about [insert topic here] – but it won’t spot the striking coincidence, the funny detail, the historical anecdote I never knew was there but which I suddenly realise is the framing device for my entire story. Jonn Elledge ["What cars did to your city, AI will do to your brain"](https://jonn.substack.com/p/what-cars-did-to-your-city-ai-will) **This series tests that fear against concrete workflows.** **Two background commitments guide me. First, _thinking with_ names a workflow in which judgement, selection, and responsibility remain mine. Second, _outsourcing_ names a workflow in which I cede those roles. This series is about the first.** **LLMs are tireless at reorganising and restating; they are weak at selecting what matters; philosophy depends on the latter.** **Worked example. I gave an LLM a rough paragraph from a paper on auditory perception. It collapsed a distinction I needed to keep. I replied: “Preserve the two‑way distinction; do not invent literature; keep premises explicit.” The next draft respected the structure but misweighted the objection. I then asked: “Weight by inferential role. Which premise does the work?” The revision aligned with my aims. The judgement in each step was mine; the model supplied options I filtered.** Something that sits behind these sorts of fears is the idea that if, by using an LLM, the LLM is doing the philosophy rather than the philosopher, then you’re not really doing it any more. And there’s something to this. In August 2025, it is fairly easy for students who want to have an LLM write an essay to get a good grade, as long as the teacher is not suspicious. They can do this pretty easily with one or two prompts, as long as they’re clear about what they want. And, yes, if that were all I did when I was using an LLM, then you’d say I’m not doing philosophy – in the same way that the student didn’t write their homework. But that’s not what I’m doing. The post I’ll put up simultaneously with this introductory one will make a start in explaining how I use LLMs and what this can tell us about thinking and working. However, this Substack should not be mistaken for a 'how to make money with AI' Substack. I’m certainly not going to tell you how to monetise LLM skills or anything like that – that’s not my wheelhouse at all. Although I will draw on technical details of LLMs, I am not a computer scientist. I have a working understanding and can separate out the gradient descent from the autoregression, but I don’t think that’s going to be particularly useful for readers at such a low level. What I am going to try to do is explain how I use them, and also explain why I find them fascinating and what I think this says about how we think and work. **What this is / is not** - A record of methods for thinking and writing with LLMs. - Not a monetisation playbook or 'how to' prompts newsletter. - Not a debate about machine consciousness; the focus is practice. I was born in the early 80s, so a lot of my childhood and teenage years were spent watching the blockbuster science‑fiction movies of the 80s and 90s. It seems to me there were three epics those movies were about: aliens, time travel, and AI. And we have that last one now! Maybe I’d be more excited if aliens landed, but I can’t understand why people aren’t more excited that we’re getting closer to AI like in the movies. Two clarifications. I am not an AI booster; I do not know whether AI will be an unalloyed good, and I do not plan to write policy takes here. Also, when I say we feel close to 'Hollywood AI', I do not mean these systems are conscious or persons, nor likely to be so in the near future. My focus here is on workflows and thinking practices. _(aside: The weird thing in philosophy, especially with technology: if somebody is writing a paper about philosophy and virtual reality, or aesthetics and virtual reality, would you expect them to know the latest models of VR headsets? In parts of aesthetics, at least, this doesn’t seem to be expected by people whatsoever. I’m also looking at how colleagues and friends of mine use LLMs. Again, it seems quite far removed from how other people might use them.)_ _Next up: Interrogating texts with an LLM._ **With one worked example and the limits of the method.** **Target audience (working notes)** - Philosophers and humanities scholars who are AI‑curious but sceptical of hype. - Writers and editors who want better thinking workflows, not hustle tips. - Research‑adjacent knowledge workers (policy, think‑tanks, academia‑adjacent). - Educators grappling with LLMs and authorship norms. - Philosophers of mind and cognitive scientists interested in the nature of thinking. \ ### Target audience (working notes) - Philosophers and humanities scholars who are AI‑curious but sceptical of hype. - Writers and editors who want better thinking workflows, not hustle tips. - Research‑adjacent knowledge workers (policy, think‑tanks, academia‑adjacent). - Educators grappling with LLMs and authorship norms. - Philosophers of mind and cognitive scientists interested in the nature of thinking. In a recent piece, Freddie deBoer expresses what I take to be a pretty common sentiment about generative AI in 2025: _%%add hyperlink to piece%%_ > a lot of people who call themselves writers appear to spend most of their days talking themselves into believing that having ChatGPT do their work is still ‘really writing.’ The Dereliction of Thought* In a recent piece, Freddie deBoer expresses what I take to be a pretty common sentiment about generative AI in 2025: %%add hyperlink to piece%% a lot of people who call themselves writers appear to spend most of their days talking themselves into believing that having ChatGPT do their work is still ‘really writing.’ I’m one of the people he’s talking about. I am a philosopher: I do a lot of writing, I use LLMs to do it, and I am pretty convinced that what I am doing is really writing. Worse, I am pretty sure that I use LLMs to think, and that this is still really thinking. This is what I intend this Substack is going to be about: AI, writing, and thinking. Why this exists In December 2022, I was miserable and bored. Although I'd recently been offered a new research position, my new institution had decided (I can only assume as a joke) to initiate every bureaucratic trick in the book to prevent me from actually starting the job. My last position had finished a few months ago, and money was starting to run out. Although I was preparing for the new position, my heart was really in it. Instead, I ended up spending a lot of my days conversing with/testing the limits of/generally antagonising ChatGPT, which had been released at the very end of November.  I’ve always been very interested in consumer technology since my dad bought me and my brothers a Sinclair ZX Spectrum +2 when I was about five. I'm the person who can't wait to help you choose your next phone or laptop and, as AI is (was) the new tech on the block, it fit the pattern. At the same time, there's a giddiness to my excitement and interest in AI that I don't remember having about any other sort of technology. I was born in the early 80s, and grew up watching the sorts of big-budget Hollywood sci-fi. And it seems to me that those sorts of movies were typically about one, or a combination of, three different subjects: aliens, time-travel, AI. So far, no Xenomorphs and no rifts in time, but we kind of have that last one now! Kind of! (footnote: hold your horses, I will explain myself in a moment). The kid inside me, the one who has seen Terminator/Blade Runner/RoboCop/Short Circuit/etc., really can't understand why people aren't more excited.  Two clarifications. First, let me qualify that 'Kind of' in that last paragraph: To be clear, I don't think that LLMs are conscious, or that they are persons, or person-like or anything like that. (footnote, I might be tempted to say that they display intelligence, but I think that can be understood as something that non-living beings exhibit as well as living ones) My position on the nature of LLMs will get clearer over the next few pieces I write. Second, don't mistake giddiness for boosterism: I do not think that this wave of AI will be an unadulterated good thing for humanity, or even an adulterated one, and I don't think that the leaders of the companies that make these products are necessarily the good guys. How worried we should be, and, what, specifically, we should be worried about, are not questions to which I have any sort of interesting answers. So for the foreseeable future I won't be saying very much at all about this area of the discourse. What I am going to write about ### **Why this exists** In December 2022, I was miserable and bored. Although I'd recently been offered a new research position, my new institution had decided (I can only assume as some sort of joke) to initiate every bureaucratic trick in the book to prevent me from actually starting the job. My last position had finished a few months ago, and money was starting to run out. Although I was preparing for the new position, my heart was really in it. Instead, I ended up spending a lot of my days conversing with/testing the limits of/generally antagonising ChatGPT, which had been released at the very end of November.  I’ve always been consumer technology since my dad bought me and my brothers a ZX Spectrum +2 when I was about five. I'm the person who can't wait to help you choose your next phone, laptop note taking app etc, as generative AI is (was) the new tech in the tech, it fit the pattern. At the same time, there's a giddiness to my excitement and interest in AI that I don't remember having about any other sort of technology. What this enthusiasm has led to is me spending a great deal of time thinking about LLMs and the other types of generative AI that have arrived in the last few years. Quite quickly I started to throw philosophy at ChatGPT, Claude, etc.: individual questions, texts by me, texts by other people, and, as of August 2025, I am happy to say that I _do_ philosophy _with_ LLMs. When I tell people this, a reasonably common response is worrying about whether these things are smart enough and about the fact that they hallucinate. Just briefly: "smart" is a funny word here, because it's something we attribute to persons or animals, not to token‑prediction machines. But let's say they are very capable of producing text that is indistinguishable at the sentence and paragraph level from real philosophy written by real people. Again, I’ll say more about this in future posts. The more interesting – at least to me – response I get is dismay. I suspect the root of this dismay is something like the following. When I say 'I do philosophy with LLMs', what I'm actually saying is 'an LLM is doing the philosophy for me'. > It feels to me like AI might do something similar for the human brain. The sort of stuff I write – factual but intended to entertain, rather than just inform – requires reading lots of stuff and seeing what grabs me. Sure, I can ask an AI to give me a bunch of facts about [insert topic here] – but it won’t spot the striking coincidence, the funny detail, the historical anecdote I never knew was there but which I suddenly realise is the framing device for my entire story. > > Jonn Elledge ["What cars did to your city, AI will do to your brain"](https://jonn.substack.com/p/what-cars-did-to-your-city-ai-will) **This series tests that fear against concrete workflows.** **Two background commitments guide me. First, _********************************************************************************************************************************************************************************************************************************************************************************************************************************************************************************************************************************************************************************************************************************************************************************************************************************************thinking with********************************************************************************************************************************************************************************************************************************************************************************************************************************************************************************************************************************************************************************************************************************************************************************************************************************************_ names a workflow in which judgement, selection, and responsibility remain mine. Second, _********************************************************************************************************************************************************************************************************************************************************************************************************************************************************************************************************************************************************************************************************************************************************************************************************************************************outsourcing********************************************************************************************************************************************************************************************************************************************************************************************************************************************************************************************************************************************************************************************************************************************************************************************************************************************_ names a workflow in which I cede those roles. This series is about the first.** **LLMs are tireless at reorganising and restating; they are weak at selecting what matters; philosophy depends on the latter.** **Worked example. I gave an LLM a rough paragraph from a paper on auditory perception. It collapsed a distinction I needed to keep. I replied: “Preserve the two‑way distinction; do not invent literature; keep premises explicit.” The next draft respected the structure but misweighted the objection. I then asked: “Weight by inferential role. Which premise does the work?” The revision aligned with my aims. The judgement in each step was mine; the model supplied options I filtered.** I am an analytic philosopher. A significant portion of my job is writing articles to submit to philosophy journals.  ** In the past I’ve written about auditory perception and the experience of time; currently most of my work is focussed on AI (insert hyperlink to paper here).**  Something that sits behind these sorts of fears is the idea that if, by using an LLM, the LLM is doing the philosophy rather than the philosopher, then you’re not really doing it any more. And there’s something to this. In August 2025, it is fairly easy for students who want to have an LLM write an essay to get a good grade, as long as the teacher is not suspicious. They can do this pretty easily with one or two prompts, as long as they’re clear about what they want. And, yes, if that were all I did when I was using an LLM, then you’d say I’m not doing philosophy – in the same way that the student didn’t write their homework. But that’s not what I’m doing. The post I’ll put up simultaneously with this introductory one will make a start in explaining how I use LLMs and what this can tell us about thinking and working. However, this Substack should not be mistaken for a 'how to make money with AI' Substack. I’m certainly not going to tell you how to monetise LLM skills or anything like that – that’s not my wheelhouse at all. Although I will draw on technical details of LLMs, I am not a computer scientist. I have a working understanding and can separate out the gradient descent from the autoregression, but I don’t think that’s going to be particularly useful for readers at such a low level. What I am going to try to do is explain how I use them, and also explain why I find them fascinating and what I think this says about how we think and work. ### What this is / is not - A record of methods for thinking and writing with LLMs. - Not a monetisation playbook or 'how to' prompts newsletter. - Not a debate about machine consciousness; the focus is practice. I was born in the early 80s, so a lot of my childhood and teenage years were spent watching the blockbuster science‑fiction movies of the 80s and 90s. It seems to me there were three epics those movies were about: aliens, time travel, and AI. And we have that last one now! Maybe I’d be more excited if aliens landed, but I can’t understand why people aren’t more excited that we’re getting closer to AI like in the movies. Two clarifications. I am not an AI booster; I do not know whether AI will be an unalloyed good, and I do not plan to write policy takes here. Also, when I say we feel close to 'Hollywood AI', I do not mean these systems are conscious or persons, nor likely to be so in the near future. My focus here is on workflows and thinking practices. _(aside: The weird thing in philosophy, especially with technology: if somebody is writing a paper about philosophy and virtual reality, or aesthetics and virtual reality, would you expect them to know the latest models of VR headsets? In parts of aesthetics, at least, this doesn’t seem to be expected by people whatsoever. I’m also looking at how colleagues and friends of mine use LLMs. Again, it seems quite far removed from how other people might use them.)_ _Next up: Interrogating texts with an LLM._ **With one worked example and the limits of the method.** **The Dereliction of Thought*** **In a recent piece %%add hyperlink to piece here%%, Freddie deBoer expresses what I take to be a pretty common sentiment about generative AI in 2025:** a lot of people who call themselves writers appear to spend most of their days talking themselves into believing that having ChatGPT do their work is still ‘really writing.’ \ I am one of those people. I am a philosopher: I do a lot of writing, I use LLMs to do it. I have convinced myself pretty damn thoroughly that what I am doing is really writing. Worse, I am pretty sure that I use them to write  I am pretty damn convinced that I use , I am someone who uses large language models (LLMs) to write. Not only that, I use them to think. Most likely, if you think that writing with LLMs isn't really writing, you're very unlikely to think that thinking with LLMs is really thinking. I started this Substack because I'm interested in questions that surround these sorts of ideas. I'm interested in authorship and creativity. I'm interested in writing. I'm interested in thinking. **This newsletter asks what changes in thinking and authorship when a philosopher collaborates with LLMs.** **This project examines authorship, workflow, and cognition in the presence of LLMs, with methods from analytic philosophy and practice from day-to-day writing.** **Why this exists** I am an analytic philosopher. A significant portion of my job is producing text for philosophy journals. **My job is reading, drafting, revising, and defending arguments for journals and talks.** In the past I’ve written about auditory perception and the experience of time; currently most of my work is focussed on AI (insert hyperlink to paper here). Since the arrival of GPT-3.5 in November 2022, I’ve been pretty much obsessed with generative AI. I've always been into tech – the sort of person who will ask you what phone model that is – and AI is the new technology on the block, so it certainly fits the pattern. At the same time, there's a giddiness to my excitement and interest in AI that I don't remember having about any other sort of technology. _[Aside: outlet for AI thoughts without burdening friends or full papers.]_ What this enthusiasm has led to is me spending a great deal of time thinking about LLMs and the other types of generative AI that have arrived in the last few years. Quite quickly I started to throw philosophy at ChatGPT, Claude, etc.: individual questions, texts by me, texts by other people, and, as of August 2025, I am happy to say that I _do_ philosophy _with_ LLMs. When I tell people this, a reasonably common response is worrying about whether these things are smart enough and about the fact that they hallucinate. Just briefly: "smart" is a funny word here, because it's something we attribute to persons or animals, not to token‑prediction machines. But let's say they are very capable of producing text that is indistinguishable at the sentence and paragraph level from real philosophy written by real people. Again, I’ll say more about this in future posts. The more interesting – at least to me – response I get is dismay. I suspect the root of this dismay is something like the following. When I say 'I do philosophy with LLMs', what I'm actually saying is 'an LLM is doing the philosophy for me'. It feels to me like AI might do something similar for the human brain. The sort of stuff I write – factual but intended to entertain, rather than just inform – requires reading lots of stuff and seeing what grabs me. Sure, I can ask an AI to give me a bunch of facts about [insert topic here] – but it won’t spot the striking coincidence, the funny detail, the historical anecdote I never knew was there but which I suddenly realise is the framing device for my entire story. Jonn Elledge ["What cars did to your city, AI will do to your brain"](https://jonn.substack.com/p/what-cars-did-to-your-city-ai-will) **This series tests that fear against concrete workflows.** **Two background commitments guide me. First, _****************************************thinking with****************************************_ names a workflow in which judgement, selection, and responsibility remain mine. Second, _****************************************outsourcing****************************************_ names a workflow in which I cede those roles. This series is about the first.** **LLMs are tireless at reorganising and restating; they are weak at selecting what matters; philosophy depends on the latter.** **Worked example. I gave an LLM a rough paragraph from a paper on auditory perception. It collapsed a distinction I needed to keep. I replied: “Preserve the two‑way distinction; do not invent literature; keep premises explicit.” The next draft respected the structure but misweighted the objection. I then asked: “Weight by inferential role. Which premise does the work?” The revision aligned with my aims. The judgement in each step was mine; the model supplied options I filtered.** Something that sits behind these sorts of fears is the idea that if, by using an LLM, the LLM is doing the philosophy rather than the philosopher, then you’re not really doing it any more. And there’s something to this. In August 2025, it is fairly easy for students who want to have an LLM write an essay to get a good grade, as long as the teacher is not suspicious. They can do this pretty easily with one or two prompts, as long as they’re clear about what they want. And, yes, if that were all I did when I was using an LLM, then you’d say I’m not doing philosophy – in the same way that the student didn’t write their homework. But that’s not what I’m doing. The post I’ll put up simultaneously with this introductory one will make a start in explaining how I use LLMs and what this can tell us about thinking and working. However, this Substack should not be mistaken for a 'how to make money with AI' Substack. I’m certainly not going to tell you how to monetise LLM skills or anything like that – that’s not my wheelhouse at all. Although I will draw on technical details of LLMs, I am not a computer scientist. I have a working understanding and can separate out the gradient descent from the autoregression, but I don’t think that’s going to be particularly useful for readers at such a low level. What I am going to try to do is explain how I use them, and also explain why I find them fascinating and what I think this says about how we think and work. **What this is / is not** - A record of methods for thinking and writing with LLMs. - Not a monetisation playbook or 'how to' prompts newsletter. - Not a debate about machine consciousness; the focus is practice. I was born in the early 80s, so a lot of my childhood and teenage years were spent watching the blockbuster science‑fiction movies of the 80s and 90s. It seems to me there were three epics those movies were about: aliens, time travel, and AI. And we have that last one now! Maybe I’d be more excited if aliens landed, but I can’t understand why people aren’t more excited that we’re getting closer to AI like in the movies. Two clarifications. I am not an AI booster; I do not know whether AI will be an unalloyed good, and I do not plan to write policy takes here. Also, when I say we feel close to 'Hollywood AI', I do not mean these systems are conscious or persons, nor likely to be so in the near future. My focus here is on workflows and thinking practices. _(aside: The weird thing in philosophy, especially with technology: if somebody is writing a paper about philosophy and virtual reality, or aesthetics and virtual reality, would you expect them to know the latest models of VR headsets? In parts of aesthetics, at least, this doesn’t seem to be expected by people whatsoever. I’m also looking at how colleagues and friends of mine use LLMs. Again, it seems quite far removed from how other people might use them.)_ _Next up: Interrogating texts with an LLM._ **With one worked example and the limits of the method.** **Target audience (working notes)** - Philosophers and humanities scholars who are AI‑curious but sceptical of hype. - Writers and editors who want better thinking workflows, not hustle tips. - Research‑adjacent knowledge workers (policy, think‑tanks, academia‑adjacent). - Educators grappling with LLMs and authorship norms. - Philosophers of mind and cognitive scientists interested in the nature of thinking. \ ### Target audience (working notes) - Philosophers and humanities scholars who are AI‑curious but sceptical of hype. - Writers and editors who want better thinking workflows, not hustle tips. - Research‑adjacent knowledge workers (policy, think‑tanks, academia‑adjacent). - Educators grappling with LLMs and authorship norms. - Philosophers of mind and cognitive scientists interested in the nature of thinking. >"a lot of people who call themselves writers appear to spend most of their days talking themselves into believing that having ChatGPT do their work is still ‘really writing.’" — Freddie de Boer * I am one of these people. * I am an analytic philosopher. * I spend my days producing philosophical texts. * I use LLMs extensively in this work. * Not only are the words and texts processed, refined, rearranged, and finessed via LLMs, but so too are the ideas. * Since the beginning of last year, all of the ideas, arguments, and theories I have developed have been done so using LLMs extensively. * I would not have produced this work without access to LLMs. * As a philosopher, I am not especially concerned with whether an idea originated from philosopher X or philosopher Y. * There is a common misconception that 21st-century philosophers are often just acolytes of particular historical figures. This is certainly not true for me. * ## Talking Myself Into Belief * The process of talking myself into believing that this is really writing completed itself a while ago. * ## Am I Kidding Myself? * It doesn't seem like I am. * It still feels like work. * It's not as though I'm typing in "do some philosophy" and then leaving the room to come back later. * Until about 2023, much of my working life was spent staring at a screen. * Now I am staring at ChatGPT, Raycast, Gemini, or several chats used concurrently instead of Google Docs. * Am I doing the philosophical work or am I just a cheat? * I don’t care much — I am interested in philosophical ideas regardless of origin. * My feelings of intense relaxation about using LLMs are related to this attitude. * Over the course of a morning, I might have several LLM chats on various topics, often working through my own ideas. * During this process, an LLM may produce a particular expression of an idea or a connection I hadn’t thought of before. * We should not underestimate these systems’ capacity to produce reasonably well-thought-through, well-grounded philosophical ideas. * This is to be expected if you consider how LLMs function. * Of course, LLMs can and do hallucinate. * Nonetheless, a particular idea generated by an LLM would not have emerged without my prompting. * Prompting well is essential for these systems to give you good ideas. * I feel a degree of ownership over all the ideas that have come out of LLMs when I’ve been using them. * ## Why This Substack * Although this feels like I'm really thinking and writing, part of the reason to start writing this Substack is to explore these ideas. * I want to work out whether I'm right. * I want to work out what it means to think and use artificial intelligence, especially LLMs, to do interesting, thoughtful work. * Another motivation for making this Substack: * Many of the people I read online and like are, if not AI doomers, then deeply dubious — brackets "filled with doob," as we used to say when I was a kid — about the value of AI to people and to the world generally. * One of them is Freddie de Boer, quoted at the top of this article. * Others include Oliver Berkman and John Elledge, who also seem, yes, not happy with the way things are going. * I am not really trying to argue against these sorts of ideas. * One hope I have is that by showing how I use LLMs, I can show people not just the possibilities. * Remember, I am not an efficiency or productivity blogger. * I also want to try and convince these people that good humanities-flavoured research, work, and writing can be done with LLMs. * A philosopher friend recently told me that, in the same way he isn’t very interested in AI-generated music or images, he wouldn’t be interested in AI-generated philosophy. * This reflects an interesting split: does philosophy need to be done by people? * My intuitions lean one way, but I want to explore and push against them to test them. * ## What This Blog Is * Although I will talk about how I use LLMs extensively, the aim of this blog is not to give weekly tips on how best to use ChatGPT. * Someone interested in replicating my techniques might learn something, but I am not a productivity blogger. * I intend to write about what interests me. * The practicalities of how I use these tools interest me. * I am a philosopher interested in the theoretical aspects of LLMs. * I am not a computer scientist, but I have a working knowledge of how LLMs work. * I can tell my gradient descent from my autoregression. * I am learning more about the technical aspects of these systems. * I am also a long-time tech enthusiast. * My first excitement over technology was for the minidisc player I bought with paper round money around 1997\. * For 25 years, I have been the person who thought hard and talked about his phone or laptop. * Coming across AI in 2023 was a continuation of this love of tech. * ## AI Arrival * AI excited me in particular. * I can’t understand why others are not as wild-eyed. * In the 80s and 90s, science fiction had three big themes: aliens, time travel, and robots. * We now have robots (AI). * We should be as excited about AI as we would be if aliens landed or a time traveller arrived. * ## What I Am Not * I am not an AI booster. * I do not think AI is unequivocally good for the world. * I do not see people like Sam Altman as prophets sent from heaven. * This is not me saying AI is wonderful. * My focus is my personal interest and fascination. * I will tell you how I do philosophy with AI, but I am not saying everyone should do this. * This works for me and nothing more. *